Evaluation of three prediction formulas of 24-hour urinary sodium excretion in Chinese residents: a systematic review and meta-analysis

Objective: To determine the appropriateness of three widely used formulas estimating 24-h urinary Na (24hUNa) from spot urine samples in the Chinese population. Design: Systematic review and meta-analysis. Setting: Literature review was conducted to identify studies for estimating 24hUNa using the Kawasaki, Tanaka and INTERSALT formulas simultaneously in PubMed, Embase and the Cochrane library databases. The mean difference (MD) and correlation coefficients (r) between measures and estimates from different formulas were assessed. Participants: Information extraction and quality assessment were performed in thirteen studies involving 8369 subjects. Results: Two studies which affected the overall robustness were excluded in the ‘leave-one-out’ sensitivity analyses. Within the final meta-analysis included eleven studies and 7197 participants, 36·07 mmol/d (95 %CI 16·89, 55·25) of MD was observed in the Kawasaki formula, and –19·62 mmol/d (95 %CI –37·37, –1·87) in the Tanaka formula and –35·78 mmol/d (95 %CI –50·76, –20·80) in the INTERSALT formula; a pooled r-Fisher’s Z of 0·39 (95 %CI 0·32, 0·45) in the Kawasaki formula, 0·43 (95 %CI 0·37, 0·49) in the Tanaka formula and 0·36 (95 %CI 0·31, 0·42) in the INTERSALT formula. Subgroup analyses were conducted to explore the possible factors affecting the accuracy of the formula estimation from three mainly aspects: population types, Na intake levels and urine specimen types. Conclusions: The meta-analysis suggested that the Tanaka formula performed a more accurate estimate in Chinese population. Time of collecting spot urine specimens and Na intake level of the sample population might be the main factors affecting the accuracy of the formula estimation.

measurement of Na in a spot urine has the potential to estimate the 24-h urinary Na (24hUNa) (7) , when assessing Na intake at the population level (8) .Currently, there are three widely used formulas estimating 24hUNa from spot urine samples among Chinese population including the Kawasaki formula (9) , the Tanaka formula (10) and the INTERSALT formula (11) .The appropriateness of the three formulas has not been systematically evaluated due to their diverse predictive results in different studies (12)(13)(14)(15)(16)(17)(18)(19) .The aim of this meta-analysis is to identify which formula could produce a more accurate estimate in the Chinese population.

Databases and search terms
The systematic review was guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses checklist (20) .A combination of terms was used to search the titles and abstracts of publications in PubMed, Embase and the Cochrane library, including 'spot urine' AND 'sodium' AND 'China' (Chinese residents).Search results were imported to EndNote X9 (Clarivate Analytics, 2019) for screening and extraction.The literature screening process is illustrated in Fig. 1.

Inclusion and exclusion criteria
All eligible studies need to meet the criteria below: (1) Type of participants.Participants included healthy or high-risk people of any age from any region of China; and (2) Type of outcome measures.Studies reported mean Na intake by estimation of spot urine and 24-h urine collection specimens at the same time; and (3) Method of validation.The Kawasaki, Tanaka and the INTERSALT formulas were used to report the estimates and measures (mean ± SD) at the same time; and (4) Language.All studies were published in English; and (5) Studies were eligible if published before November 2022.
Studies were excluded if they did not provide the measure of 24hUNa and estimates from spot urine using the three formulas simultaneously.

Study screening and data extraction
All titles and abstracts of included studies were initially screened by two reviewers (ZQ, ST), and the full text of potentially relevant articles was further reviewed.Then, the two reviewers independently extracted relevant data as follows: the first author, publication year, area studied, study population, sample size, female proportion, mean age (age range), fasting or not, type of spot urine sample, measured 24hUNa (mean ± SD), the formula used, estimated 24hUNa (mean ± SD) and correlation coefficients (r).
Reviewers resolved disagreements through consulting senior researchers (ZL) if necessary, during the study screening, review and data extraction.Details of the included thirteen studies are summarised in Table 1.

Statistical analysis
Pooled mean difference (MD) and pooled r-Fisher's Z of the three formulas were calculated using the Sidik-Jonkman method (21) for random-effect models.Sensitivity analyses were carried out by the 'leave-one-out' method, which removed one study at a time to check the robustness of the result.If the pooled results showed relatively large biases when a study was excluded, it suggests that the study had a significant impact on the overall results of metaanalysis, leading to its exclusion.Heterogeneity between studies was reflected by I 2 tests.Subgroup analyses were also implemented to detect potential sources of heterogeneity: (1) population subgroup (healthy or high-risk population), (2) Na intake subgroup (more than 10 g/d group or no more than 10 g/d group) and (3) urine sample type subgroup (casual spot urine, morning urine or fasting morning urine).The meta-analysis was conducted at the population level using software RevMan 5. Na excretions were converted into mmol/d for consistency, using the following equations: 1 mmol Na = 23 mg Na and 1 mg Na = 2•54 mg salt (1) .Spearman correlation coefficients were converted to Pearson correlation coefficients for consistency using the following equations: r = 2×sin(r s × π/6) (22) .Correlation coefficients (r) reported in each study were converted into Fisher's Z using the following specific formula (23) .The conversion formula between the correlation coefficient and Fisher's Z is as follows:

Quality assessment
Study quality was independently assessed by two reviewers (ZQ, ST) using a modified tool for evaluating dietary intake validation studies (24) .The tool rates the studies through five domains on a scale from 0 (poorest quality) to 7 (highest quality), with the following interpretations: excellent if the score ≥5•0; good if ≥3•5 and <5•0; acceptable or reasonable if ≥2•5 and <3•0; and poor if <2•5.
The quality assessment domain was modified to facilitate the estimated 24hUNa by spot urine (1) , through the consideration of variables as below: (1) sample, with a maximum of 1 point: 0•5 point when the sample size was of more than 100 individuals; and 0•5 point allocated when the sample was homogeneous for sex, age and population; otherwise, sample was considered non-homogeneous and given a score of 0; (2) statistics.A maximum of 3 points was allocated; 1 for MD between estimated and measured values; 0•5-1 point according to the correlation used (r, ICC); 0•5-1 according to the agreement used Bland-Altman plots or rate of misclassification.(3) Data collection with a maximum of 1 point: 0•5 point if verbal or written instructions were conducted to collect urine to the participants; plus 0•5 when spillage or missed voids were assessed post-collection.(4) Seasonality.Addition of 0•5 point only when considered in the validation design.
(5) Supplements.Addition of 1•5 points when the validation study considered supplements.The results of quality assessment are shown in Table 2.

Literature selection
One hundred seventy-nine articles were retrieved initially, of which 142 articles were excluded, leaving thirty-seven studies for full-text review.Finally, thirteen studies were included in this review, and eleven studies were conducted in the meta-analysis.The study selection process is presented in Fig. 1.

Sensitivity analyses
As shown in Fig. 2, sensitivity analysis using the 'leave-oneout' approach excluded Yi Zhao et al.'s study (17) and Wenxia Ma et al.'s study (13) which had great impact on the robustness of the overall results.Table 3 shows the impact of the two excluded studies on the meta-analysis.

Overall result
The final meta-analysis as shown in Fig. 3 (31) , the correlation coefficient converted from r-Fisher's Z showed that the Tanaka formula (0•41) performed a moderate correlation, while the Kawasaki formula (0•37) and the INTERSALT formula (0•35) performed a weak correlation.Through the comparison of the statistical significance of the difference between two independent correlation coefficients, the Tanaka formula had a higher correlation than the Kawasaki formula (0•41 v. 0•37, P = 0•01) and the INTERSALT formula (0•41 v. 0•35, P < 0•01).

Subgroup analysis
For those consuming over 10 g/d, the Kawasaki formula indicated a MD of 22 Fig. 2 Sensitivity analyses in use of the 'leave-one-out' method: pooled estimates were from random-effects models with removing one study at a time Sensitivity 1, including all available thirteen studies; Sensitivity 2, excluded one study (17) that had a large impact on the overall results from thirteen studies; Sensitivity 3, excluded two studies (13,17) that had large impacts on the overall results from thirteen studies.The detailed results are summarised in Fig. 5.

Discussion
Using spot urine samples could be a convenient alternative to 24-h urine collections for monitoring Na intake at the population level (26,32) .The three formulas in the metaanalyses all showed moderate correlations (0•36-0•43) between the estimated and measured 24hUNa, slightly lower than those observed in the Italian population (0•62-0•70) (33) and slightly higher than the Portuguese population (0•25-0•36) (34) .In a diverse population from eleven countries (35) , the Kawasaki formula was considered a reliable alternative for estimating 24hUNa while the INTERSALT formula was seen as an acceptable alternative for monitoring Na intake/excretion in the French Fig. 3 Forest plot of mean difference (MD) between measures and estimates from different formulas population (36) .Several foreign validation studies (37,38) also identified that the INTERSALT formula which derived from data across fifty-two centres in thirty-two countries perform better under complex ethnic compositions.This metaanalysis in the Chinese population suggested the Tanaka formula showed a less bias (- Subgroup analysis identified two key factors that might influence the accuracy of formula estimates.First, Na excretion exhibited a diurnal variation (39) , implying that the time of urine sample collection could impact the overestimation or underestimation of 24-hour urinary sodium excretion (24hUNa) (40,41) .The Kawasaki formula initially was established based on fasting second morning urine specimens (9) , which had a lower Na concentration (27,42) , Therefore, the Kawasaki formula produced higher estimates when collecting casual spot urine specimens, and the Tanaka and the INTERSALT formula produced lower estimates when fasting morning urine was collected.Second the level of Na intake appears to play a role in estimation accuracy.Numerous studies (12,18,19,29,39,42) have demonstrated that prediction formulas are more likely to overestimate in populations with lower salt intake and underestimate in those with higher intake.This was consistent with the trends presented in the subgroups of urinary Na intake levels.
The 24hUNa estimates by the Kawasaki formula tended to be overestimated, those by the Tanaka and INTERSALT formulas were inclined to be underestimated, consistent with trends identified in Brazilian (43) , Iranian (33) and Indian populations (37) .More accurate estimation results might be achieved by closely approximating the conditions under which the formula was established.Therefore, the estimates derived from the Kawasaki formula in fasting morning urine group (24•17 mmol/d) and in the group with salt intake group over 10 g/d (22•14 mmol/d) are more accurate compared with those obtained by the other two formulas.
Among the studies excluded from the sensitivity analysis, one focused on the elderly high-risk stroke population in rural area of Shaanxi province, and the other on elderly hypertensive patients in a rural area of the Ningxia Hui Autonomous Region.Both studies involved high-risk elderly populations with lower urine volumes and higher urine Na concentrations, resulting in significantly higher estimates compared with the measures.Although these two studies were not included in the final metaanalysis, there is a need to validate more accurate formulas in more diverse populations or in larger epidemiological studies within China.
There were two limitations in this study.First, the metaanalysis mainly evaluated the validity of the Kawasaki formula, the Tanaka formula and the INTERSALT formula, involving only 8369 participants from thirteen studies in fifteen provinces of China, which might limit its extrapolation.Second, correlation might not be the best measure to assess the validity in the current context of monitoring and evaluating public health programmes for population salt reduction (44) .A more comprehensive and in-depth evaluation of different formulas should be implemented in the future.

Conclusion
The meta-analysis suggested that the Tanaka formula estimates exhibited less bias and higher correlation in the Chinese population.The time of collecting spot urine specimens and Na intake level of the sample population might be the main factors affecting the accuracy of the formula estimation.

Fig. 4
Fig. 4 Forest plot of correlation (r) between measures and estimates from different formulas

Table 1
Characteristics of included studies Studies included in meta-analysis (MD: n 11; r: n 10) Identification Screening Eligibility Included Fig. 1 PRISMA flow diagram of the screening procedure followed to identify eligible studies.PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses; MD, mean difference; r, correlation coefficient

Table 2
Quality assessment of the included studies

Table 3
Sensitivity analyses of mean difference (MD) in different formulas