Background Numerous cohort studies with sodium intake measured by gold standard method (i.e. non-consecutive 24h urinary sodium excretion) identified a positive linear association of sodium intake with cardiovascular disease (CVD) risk. However, several studies using formulas to estimate sodium intake from spot urine reported opposite findings. We aim to evaluate whether and how these formulas lead to these different findings.

Methods UK Biobank is a prospective cohort study with over 500,000 participants aged 40-69 years in the UK recruited during 2006-2010 and followed up. Random spot urine samples were collected and sodium concentration was measured. Sodium intake was estimated by INTERSALT, Kawasaki, and Tanaka formulas, respectively. The primary outcomes included CVD events and deaths. Hazard ratios (HRs) were estimated by Cox proportional-hazard model adjusted for multiple covariates. Quartiles and penalised splines of estimated sodium intakes were assessed to test both linear and non-linear relationships. We also fixed the sodium concentration with sex-specific mean and re-ran the above-mentioned analyses to assess whether and how formulas themselves impacted these associations. All analyses were performed on R v4.2.2.

Results 435,336 participants were included. 44,268 CVD events and 3,251 CVD deaths occurred during a median follow-up of 12 years. The mean estimated sodium intake was 142.5 (SD=34.6), 178.17 (51.75), and 146.88 (32.79) mmol/day from INTERSALT, Kawasaki and Tanaka formulas. For CVD incidence, linear inverse associations were observed for INTERSALT and Tanaka estimates (HRs [95% CIs]: 0.9 [0.83, 0.97] and 0.93 [0.89, 0.97] for every 50 mmol reduction in estimated daily sodium intake; P-linear=0.0047 and 0.0021), and U-shape association for Kawasaki (P-nonlinear=0.0026). When the sodium concentration term was fixed in the formulas, inverse association was seen for all formulas (HRs [95% CIs]: 0.86 [0.77, 0.95], 0.96 [0.93, 0.99] and 0.94 [0.89, 0.99] for INTERSALT, Kawasaki and Tanaka; P-linear=0.0054, 0.0166 and 0.0188). For CVD mortality, no association was observed for any formula, but a non-linear association for INTERSALT after fixing the sodium concentration term (P-nonlinear=0.0287).

Conclusion A nonlinear or inverse linear association with CVD incidence was found for formula estimates of sodium intake, and no association was found with CVD mortality. The observed relationships were problematic as formulas were associated with outcomes independently of sodium concentration. Formulas to estimate sodium intake from spot urine samples should not be used in research investigating associations of sodium intake with CVD outcomes, as they are associated with the controversy of benefits from sodium reduction and may hinder implementation of relevant policies.