Global and regional reference curves for pain across the lifespan in 6.1 million individuals in 118 countries

The logistical achievement is substantial: pooling individual-level data from 6.1 million people across 118 countries, with systematic harmonization of heterogeneous survey instruments and hierarchical modeling that accounts for the nested data structure. For its descriptive aim — mapping who reports pain, where on the body, and how that varies by age and geography — the design is appropriate and well-executed.

The design cannot support the paper's interpretive claims. The inverted-U trajectory has three alternative explanations (survivorship bias, cohort effects, reporting diffusion) that are acknowledged in the limitations but not tested or bounded, while the results, discussion, and abstract treat the pattern as a lifecourse phenomenon. The boldest comparison — a 30.6-percentage-point difference in late-life pain between the lowest- and highest-development countries — is estimated at ages where very few people in low-development countries survive. Data coverage is heavily skewed: western Europe contributes about 19 data sources per country; sub-Saharan Africa contributes about 2. The population attributable fractions use causal language ("attributable to") for observational associations with no causal identification strategy, and the income variable (within-study quintiles) measures different things in rich and poor countries by construction.

The analytic framework — mixed-effects logistic regression with age splines, random effects for study and region, cross-validated spline selection, Benjamini-Hochberg correction for multiple comparisons — is technically sound and well-suited to pooling binary outcomes from hierarchically structured data. Effect sizes with confidence intervals are reported consistently, and the sensitivity analyses are thorough.

The problems are interpretive. The tight confidence interval on the headline HDI estimate (26.9 to 34.3 percentage points) reflects partial pooling, which borrows strength from data-rich strata and narrows uncertainty in data-sparse cells. The estimate looks more precise than the underlying data support at the specific age-by-development-level combination where the claim is made. Survey year was omitted from the development-level models because it correlated with the development index, meaning the analysis cannot separate national development from temporal trends in pain reporting over the 35-year data window. The population attributable fractions require causal risk ratios to be valid, but the study provides only cross-sectional associations confounded by occupation, comorbidities, and healthcare access. Natural cubic splines are constrained to be linear beyond their boundary knots, so the late-life portions of the curves — where the largest claims are made — are partly driven by model structure rather than data.

The basic measurement — self-reported pain at 11 anatomical sites, harmonized across studies with different recall periods, question wording, and languages — is reasonable for descriptive epidemiology. The harmonization procedure adjusted for recall period and case definition and evaluated five additional methodological features. But nearly every interpretive claim stretches the construct beyond what it measures.

Cross-sectional age patterns are not lifecourse trajectories. The composite "any bodily pain" measure rises monotonically with age while individual body sites show the inverted U, consistent with older adults redistributing pain reporting from single sites to diffuse whole-body pain rather than experiencing less pain. The prevention-window claim chains two unsupported steps: cross-sectional patterns are assumed to reflect individual accumulation, and accumulation timing is assumed to determine optimal prevention timing. The comparison with the Global Burden of Disease framework treats a divergence between two measurement approaches (self-report versus clinical diagnosis) as evidence that one is wrong, when the divergence is expected from the definitional mismatch alone. The "fundamentally different drivers" conclusion in poorer regions cannot be distinguished from "same drivers, worse measurement" — if BMI data are less accurate or if relevant tobacco exposures (hookah, smokeless tobacco) are not captured by the "current smoker" variable, the attributable fractions would shrink without the underlying causes differing. No measurement invariance analysis tests whether pain questions function equivalently across languages and cultural contexts.

The descriptive findings are presented clearly. The problems arise when the paper moves from description to interpretation. The abstract, results, and discussion use "trajectory" and "accumulation" language for cross-sectional age patterns, building a prevention-timing recommendation on a pattern that has multiple non-lifecourse explanations the study cannot distinguish. The Global Burden of Disease comparison is framed as a contradiction ("Contrary to existing estimates..."), but the two frameworks measure different constructs — self-reported pain at any body site versus clinically defined conditions — and disagreement between them is expected. The paper's preferred explanation (condition-based approaches miss undiagnosed pain in poorer countries) is plausible but equally consistent with self-report capturing a broader, less specific construct, or with cross-cultural differences in pain reporting inflating self-report in some settings. The "fundamentally different drivers" conclusion overstates what low explained variance from three variables can show. A low attributable fraction does not distinguish between different causes operating and the same causes being measured with less precision.

Analysis code is publicly available on GitHub. Conflict and funding disclosures are complete. The limitations section addresses the cross-sectional design, uneven coverage, measurement heterogeneity, and survivorship effects. Two things are missing: the paper does not report how many participants aged 80+ from low-development countries contributed to the headline estimate, and the income variable's non-comparability across regions is not flagged as a limitation. The individual-level data cannot be shared due to data-use agreements, which prevents independent replication.

Citations are appropriate for the descriptive and methodological claims. The GBD comparison cites four specific GBD publications and two critical commentaries, and the paper's characterization of the methodological differences is transparent. The candidate explanations for why pain drivers differ in poorer countries (occupational load, limited rehabilitation, infections, vitamin D deficiency) each cite a single paper, none specific to the populations where the difference is largest. The paper does not cite the literature on cultural variation in pain reporting — studies documenting that reporting thresholds and willingness to report pain differ across cultures and languages — which is directly relevant to its central finding. The literature on survivorship bias in cross-sectional aging studies is also absent.

Funding is entirely from public and philanthropic sources with no industry involvement. The lead investigators have no disclosed industry ties. Four coauthors from a contributing neuroimaging consortium declare pharmaceutical industry relationships, but none are in therapeutic areas related to pain epidemiology and none are among the analysts. The study makes no treatment recommendations and has no commercial implications.