Infectious Disease · Medical Literature
[Correspondence] How definitions shape obesity-attributable death estimates
Abstract
Solja T Nyberg and colleagues1 report a consistent dose–response association between adult obesity and severe infections across Finnish cohorts and UK Biobank, and estimate that worldwide 8·6% (95% CI 6·6–11·1) of infection-related deaths in 2018, 15·0% (12·8–17·4) in 2021, and 10·8% (8·6–13·6) in 2023 might be attributable to obesity. These estimates are important for prevention policy; however, their interpretation depends on modelling choices and assumptions.
Clinical relevance
Do the methodological and definitional assumptions underlying the obesity-attributable infection death estimates reported by Nyberg et al. materially affect the interpretation and policy applicability of those population attributable fractions (PAFs)? __SNAPSHOT__ {"studyDesign":"Correspondence (letter) responding to a multicohort prospective study with global burden modelling; not an original data study. The index study by Nyberg et al. used pooled prospective cohort data (two Finnish cohorts + UK Biobank) with PAF modelling applied to Global Burden of Disease (GBD) estimates.","setting":"Commentary published in The Lancet; index study conducted in Finland and the United Kingdom with global PAF extrapolation across 204 countries using GBD 2021 data.","sampleSize":"Not applicable (correspondence). Index study: Finnish cohorts n=67,766; UK Biobank n=479,498; combined analytical sample approximately 547,264 adults.","population":"Adults without a prior history of severe infection-related hospitalisations at baseline. Finnish cohorts: mean age 42.1 years (SD 10.8), 73.1% female. UK Biobank: mean age 57.0 years (SD 8.1), 54.4% female.","intervention":"Exposure of interest: adult obesity defined by BMI ≥30.0 kg/m², subclassified as class I (30.0–34.9), class II (35.0–39.9), and class III (≥40.0 kg/m²), assessed at a single baseline measurement.","comparator":"Healthy weight (BMI 18.5–24.9 kg/m²) as the reference category in Cox proportional hazards models.","primaryOutcome":"Severe infection, defined as hospital-treated infectious disease or infection-related mortality. PAF of infection-related deaths globally attributable to obesity, estimated for 2018 (pre-COVID-19), 2021 (pandemic), and 2023 (post-pandemic).","followUp":"Mean follow-up 14.1 years (SD 3.1) in Finnish cohorts (registry data through 2016); mean 12.6 years in UK Biobank (registry data through 2022).","fundingSource":"Wellcome Trust, Medical Research Council, and Research Council of Finland (index study). Funding source for the correspondence itself: not reported."}
Key findings
This correspondence does not report new primary data or statistics. Its central argument is that the PAF estimates generated by Nyberg et al.—8.6% (95% CI 6.6–11.1) of infection-related deaths in 2018, 15.0% (95% CI 12.8–17.4) in 2021, and 10.8% (95% CI 8.6–13.6) in 2023—are sensitive to modelling choices and definitional assumptions that the authors of the correspondence contend warrant explicit acknowledgment. The index study's primary effect estimates, against which these PAFs were derived, showed that class III obesity (BMI ≥40 kg/m²) was associated with a pooled HR of approximately 2.69–3.07 for combined fatal and non-fatal severe infection depending on cohort. Any obesity (classes I–III combined) carried a pooled HR of 1.7 (95% CI 1.7–1.8) for fatal or non-fatal severe infection. The PAF calculations relied on (1) hazard ratios from Northern European and UK cohorts applied to 204 countries globally, (2) GBD 2021 obesity prevalence estimates, and (3) the BMI ≥30.0 kg/m² threshold as the sole definition of obesity. Notably, the PAF is sensitive to the choice of counterfactual: published methodological literature demonstrates that shifting the reference BMI category can alter obesity-attributable death fractions substantially (e.g., from 0.7% to 4.8% in prior mortality analyses depending on reference BMI selected). Geographic heterogeneity in the index study was substantial—PAFs ranged from approximately 4.1% (South Asia) to 22.5% (North Africa and Middle East) in 2023—suggesting that uniform application of European-derived HRs to low- and middle-income settings may introduce systematic bias. NNT/NNH: not calculable from correspondence alone. Absolute death counts from the index study: approximately 0.6 million (95% CI 0.5–0.7 million) obesity-attributable infectious deaths in 2023 out of 5.4 million (95% CI 4.6–6.5 million) total infectious deaths.
Practice implications
This correspondence raises methodologically important cautions about how population attributable fraction estimates for obesity and infection mortality should be read and used by policymakers and clinicians. The index study by Nyberg et al. established a robust, dose-response association between obesity and severe infection risk across two geographically and demographically distinct cohorts, with consistency across pathogen types and adiposity measures. However, the PAF estimates—which extrapolate Northern European hazard ratios to 204 countries using GBD data—should be interpreted as modelling approximations rather than fixed epidemiological constants. Key uncertainties include: (1) whether HRs derived from predominantly female, Northern European cohorts are transportable to populations with different obesity phenotypes, infectious disease burdens, and health-system characteristics; (2) the sensitivity of PAF estimates to the definitional threshold for obesity (BMI ≥30 vs. alternative cutoffs); (3) the use of single baseline BMI measurements, which do not capture weight change over follow-up; and (4) residual confounding by socioeconomic status, comorbidities, and immunosuppressive medications. Two infection categories—HIV and tuberculosis—did not show increased risk with obesity, likely due to reverse causality, underscoring that the relationship is not uniform across all pathogens. The finding that PAFs nearly doubled during the COVID-19 pandemic (15.0% in 2021 vs. ~8.6% pre-pandemic) reflects both true biological interaction and pandemic-era changes in GBD mortality denominators. Only the abstract of the correspondence is available for review; the full text of the specific methodological critiques raised by Boulares and Bragazzi is not accessible for detailed appraisal. __LIMITATIONS__ 1. This is a correspondence, not an original research study; it provides critique without new data, and the specific methodological arguments cannot be fully evaluated without access to the full text. 2. The index study used single-point baseline BMI, which does not reflect longitudinal adiposity trajectories or clinically meaningful weight change during follow-up. 3. Finnish and UK Biobank cohorts are predominantly European, limiting generalizability to populations with different obesity-infection risk profiles, particularly in low- and middle-income countries where most infectious disease deaths occur. 4. The Finnish cohort was 73.1% female, introducing potential sex-selection bias in effect estimates applied to global all-sex populations. 5. PAF calculations assume a causal, unconfounded relationship between obesity and infection death—an assumption that cannot be fully verified in observational data. 6. The choice of counterfactual (BMI 18.5–24.9 kg/m² as 'healthy weight') is a definitional decision that materially affects the magnitude of PAF estimates; alternative reference categories would yield different results. 7. Hazard ratios used for global burden modelling were derived from high-income, predominantly White European cohorts and applied uniformly across 204 countries with highly variable obesity prevalences, infectious disease patterns, and healthcare access. 8. Residual confounding by immunosuppressive drugs, diabetes severity, socioeconomic deprivation, and other comorbidities cannot be excluded. 9. The 2021 PAF estimate (15.0%) conflates pandemic-specific COVID-19 biology with the general obesity-infection relationship, limiting its use as a baseline for non-pandemic prevention planning. __BOTTOM_LINE_CATEGORY__ potentially_practice_informing __BOTTOM_LINE__ The Nyberg et al. multicohort study establishes a credible, dose-response association between adult obesity and severe infection risk across diverse pathogens, but the global PAF estimates (8.6–15.0% of infection deaths attributable to obesity) rest on methodological assumptions—particularly the extrapolation of European hazard ratios across 204 countries and sensitivity to the obesity definition threshold—that warrant caution when translating these figures into regional or national policy. Clinicians should recognize obesity as a meaningful, modifiable risk factor for severe infections warranting integration into vaccination, weight management, and infection prevention counseling, while policymakers should treat the absolute PAF figures as order-of-magnitude estimates rather than precise targets pending country-specific validation.