This study was conducted on the basis of the CNSR III study (China National Stroke Registry III), which was a nationwide, hospital-based, prospective cohort study enrolling patients with acute ischemic cerebrovascular events between August 2015 and March 2018 in China. Patients were eligible if they met the following criteria: age 18 years or older; diagnosis within 7 days of the index event of ischemic stroke or TIA. The design, rationale and baseline patient characteristics of CNSR III study have been published previously (16). ¹⁵ A total of 201 study sites participated in this study with a median number of patients recruited as 55 (minimum, 3; maximum, 321). Among the 15,166 patients in the registry, 7,353 were analyzed after excluding patients with missing serum phosphate value (n = 7,288), missing serum creatinine or lipid value for key covariates (n = 312) and lost during the 1-year follow-up (n = 213; Figure 1).

The CNSR III study was approved by ethics committee at Beijing Tiantan Hospital (IRB approval number: KY2015-001-01), and written informed consent was obtained from patients or their legally authorized representatives. The study complied with the principles of the Declaration of Helsinki.

Baseline information of included patients, including demographics, vascular risk factors, important laboratory data, treatment and complications were collected by trained research coordinators at each study center. Fasting blood samples were drawn within 24 h of admission and assessed as baseline laboratory data. Risk factors contained stroke history, hypertension, diabetes mellitus, hypercholesterolemia, coronary heart disease, previous or current smoking, and heavy alcohol consumption. Hypertension, diabetes mellitus, and hypercholesterolemia were defined according to (1) documented or self-reported history or (2) receiving medication for corresponding diseases or (3) clinical or laboratory examination (systolic blood pressure ≥140 mmHg or diastolic blood pressure ≥90 mmHg on repeated measurements for a diagnosis of hypertension (17), fasting glucose level ≥126 mg/dl or 2-h plasma glucose ≥200 mg/dl during oral glucose tolerance test (75 g) or random plasma glucose ≥200 mg/dl in persons with symptoms of hyperglycemia or hyperglycemic crisis for diabetes mellitus (18), total cholesterol>240 mg/dl or serum triglyceride ≥200 mg/dl or low-density lipoprotein cholesterol ≥160 mg/dl or high-density lipoprotein cholesterol ≤ 40 mg/dl for dyslipidemia (19)), or (4) new diagnosis at discharge. Heavy alcohol consumption was defined as consuming ≥2 standardized alcohol drinks per day. Estimated glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration creatinine equation with adjusted coefficient of 1.1 for the Asian population (20, 21).

Fasting blood samples were drawn within 24 h of admission, and serum phosphate levels were assessed with an ammonium molybdate assay using unfrozen samples in each center. Briefly, the phosphate ions react with ammonium molybdate and then reduced to blue molybdenum which is finally colorimetric measured.

Patients were followed up over telephone at 12-month after disease onset by trained research coordinators who were blinded to baseline clinical status. Data on clinical outcomes were collected. We defined adverse clinical outcomes as recurrent stroke, all-cause mortality, and poor functional outcome. Recurrent stroke includes ischemic stroke, intracranial hemorrhage, and subarachnoid hemorrhage. Composite end point was comprised of recurrent stroke, myocardial infarction, other ischemic vascular events, and all-cause mortality. Poor functional outcome is defined as modified Rankin Scale (mRS) score of 3 to 6 [mRS score ranges from 0 (no symptoms) to 6 (death)].

Proportions were used to describe the categorical variables; means with SD or median with the interquartile range (IQR) were used for continuous variables. We used χ2 test for categorical variables; 1-way analysis of variance or Kruskal-Wallis test were adopted for continuous variables. Univariate and multivariable Cox regression models were performed to estimate the association between serum phosphate and all-cause mortality, recurrent stroke and composite end point, with third quartile of serum phosphate as reference group (14), while univariate and multivariable logistic regression models were performed to estimate the association between serum phosphate and poor functional outcome. Odds ratios (ORs) or hazard ratios (HRs) with their 95% CIs were reported. Variables included in the multivariable model were selected based on baseline characteristics differences between different quartile groups or based on previous studies. These variables were age, sex, history of stroke, hypertension, diabetes mellitus, hypercholesterolemia, coronary heart disease, current or previous smoking, heavy drinker, body mass index (BMI), National Institutes of Health Stroke Scale (NIHSS) score at admission, hemoglobin, serum calcium, serum potassium, serum albumin, estimated glomerular filtration rate, serum creatinine, total cholesterol, triglycerides, mRS at discharge, antihypertensive drugs, lipid-lowering drugs, hypoglycemia drugs, and pneumonia during hospitalization. In addition, we further explored the pattern of association between serum phosphate levels and risk of stroke outcomes using a logistic regression model with restricted cubic splines for serum phosphate adjusting for covariates with 5 knots (at the 5th, 25th, 50th, 75th and 95th percentiles). Furthermore, C statistics, net reclassification index, and integrated discrimination improvement were used to evaluate the incremental prognostic value of serum phosphate levels beyond conventional risk factors (22). All analyses were conducted with SAS version 9.4 software (SAS institute), and 2-tailed P-values of <0.05 were considered to be statistically significant.

Compared to patients excluded (n = 7,813), the patients included in this analysis were slightly older and more likely to have diabetes history, to be a heavy drinker, and to have lower mRS score at discharge (Supplementary Table 1). Baseline characteristics of included patients are summarized in Table 1. Of the total 7,353 patients, the mean age was 62.5 years, and 68.6% were men. Patients with higher quartiles of serum phosphate were younger, and there were more females, more patients with history of diabetes mellitus, hypercholesterolemia while less smokers. They also had slightly higher BMI and lower NIHSS score. The levels of serum calcium, serum potassium, serum albumin, eGFR, total cholesterol, low-density lipoprotein cholesterol and triglycerides increased along with the levels of serum phosphate, while the levels of hemoglobin and serum creatinine decreased. History of stroke, hypertension, coronary heart disease, alcohol consumption, and level of high-density lipoprotein cholesterol were not significantly different among the quartiles.

The 1-year incidences of clinical outcomes are shown in Table 2. The 1-year rates of recurrent stroke, composite endpoint and poor functional outcome were lowest in third quartile group (p = 0.03 for recurrent stroke; p = 0.03 for composite end point; p < 0.0001 for poor functional outcome). There was no significant difference in all-cause mortality among groups (p = 0.06). In the lowest serum phosphate quartile, the incidence rates of outcomes including all-cause mortality, recurrent stroke, composite endpoint, and poor functional outcomes were 4.4, 10.8, 11.6, 17.6%, respectively.

Crude and adjusted ORs or HRs with 95% CIs of serum phosphate levels for adverse clinical outcomes are presented in Table 3. Compared with the third quartile of phosphate (1.06–1.20 mmol/L), the adjusted ORs/ HRs (95% confidence interval) of the lowest quartile (<0.94 mmol/L) were 0.98 (0.67–1.42) for all-cause mortality, 1.31 (1.05–1.64) for stroke recurrence, 1.26 (1.02–1.57) for composite end point, and 1.27 (1.01–1.61) for poor functional outcome, and the adjusted ORs/HRs of the highest quartile (≥1.20 mmol/L) was 1.28 (0.87–1.86) for all-cause mortality, 1.11 (0.88–1.40) for stroke recurrence, 1.09 (0.87–1.35) for composite end point, and 1.40 (1.11–1.77) for poor functional outcome. Further analyses using restricted cubic spline regression showed a U-shaped relationship between serum phosphate levels and poor functional outcome and that low serum phosphate levels were significantly associated with increased risk of recurrent stroke and composite end point (Figure 2).

In patients with recurrent stroke (n = 674), 64 (9.5%) patients had hemorrhage stroke (Supplementary Table 2). Analyses for association between serum phosphate levels and 1-year hemorrhage stroke did not show statistical significance among groups (Supplementary Table 3).

Further analysis for association of serum phosphate levels with one-year all-cause mortality and poor functional outcome in patients without recurrent stroke (n = 6,679) did not show statistical difference between serum phosphate quartiles after adjusting for age, sex and other covariates. (Supplementary Table 4).

We further investigated whether adding serum phosphate to conventional risk factors improved the risk prediction of stroke recurrence, composite endpoint and poor functional outcome (Supplementary Table 5). Adding serum phosphate to a model containing conventional risk factors significantly improves risk reclassification for stroke recurrence (categorical net reclassification index was 8.9%, p = 0.03), composite end point (8.7%, p = 0.03) and poor functional outcome (14.3%, p < 0.001).