All SPRINT anonymized data can be found at the National Heart, Lung and Blood Institute (NHLBI) Biologic Specimen and Data Repository (https://biolincc.nhlbi.nih.gov/home/).

This study was a secondary analysis of the SPRINT trial. As mentioned above, SPRINT was designed to investigate the beneficial effect of intensive BP control in reducing cardiovascular events, as compared with standard BP control. Details of the trial have been discussed elsewhere [15]. Briefly speaking, the SPRINT trial enrolled 9361 participants between November 2010 and March 2013. The participants were aged over 50 years and had an average SBP of 130 to 180 mmHg, along with an increased risk for CVD. This was defined as having at least one of the following: clinical or subclinical CVD other than stroke, a 10-year Framingham risk score (FRS) of 15% or higher, being aged 75 years or older, or having an estimated glomerular filtration rate (eGFR) of 20 to 60 mL/min/1.73 m${}^2$.

Major exclusion criteria for the trial included diagnosed diabetes, prior stroke, severe chronic kidney disease (CKD) (eGFR 20 mL/min/1.73 m${}^2$), congestive heart failure (ejection fraction 35%), or dementia. For the purpose of this study, only participants who received intensive blood pressure treatment were included.

This study was approved by the Institutional Review Board of each clinical site, and all participants provided informed consent.

Baseline demographic data, including age, gender, and race were self-reported at randomization. Medical histories were recorded in a document called self-administered baseline history. As for anemia status, participants were asked if they had ever been told by a physician that they had anemia or low blood count. If they answered yes, then they were recorded as positive for anemia. The same kinds of questions were asked to collect other medical histories, including cancer, smoking status, and medication usage. In the SPRINT trial, prior CVD was defined as a history of clinical or subclinical CVD, including acute coronary syndrome, carotid revascularization, coronary revascularization, more than 50% stenosis of carotid/coronary/lower extremity artery, or a coronary artery calcium score of 400 or higher. Chronic kidney disease was defined as an eGFR of less than 60 mL/min/1.73 m${}^2$. The eGFR was calculated using the modification of diet in renal disease 4-component equation [16]. BP was measured using an automated device (Omron-HEM-907, OMRON HEALTHCARE Co., Ltd., Kyoto, Japan) following standard procedures. The mean of 3 automated cuff readings was used for analysis. Laboratory specimens were obtained at baseline and stored at the University of Minnesota for estimation of serum markers, including blood glucose, serum creatinine, and lipid profiles.

The primary outcome of our study was a composite of cardiovascular events, including nonfatal myocardial infarction, acute coronary syndrome not resulting in myocardial infarction, stroke, acute decompensated heart failure, or death from CVD. The definition of our primary outcome is consistent with the one used in SPRINT. Our secondary outcome was all-cause death, which is also a prespecified secondary outcome in SPRINT. All outcome events were reviewed and confirmed by experienced physicians who were blinded to the treatment assignment.

All statistical analyses were performed using R version 3.6.2 (https://www.r-project.org/; R Foundation for Statistical Computing, Vienna, Austria). A p value less than 0.05 was considered to be statistically significant. Participants in the intensive BP treatment group were divided into 2 groups based on their anemia status. The baseline characteristics of the 2 groups were compared using appropriate statistical tests. Continuous variables were compared using the Wilcoxon rank sum test or 2-sample t-test, while categorical variables were compared using Pearson’s Chi-squared test or Fisher’s exact test. Cox proportional hazards models were used to compare the incidence of composite cardiovascular events and all-cause death between the 2 groups. HRs with 95% confidence intervals (CIs) were calculated, with participants without anemia serving as the reference group.

Since this was a secondary analysis of SPRINT, baseline characteristics may not be balanced between the 2 groups. One approach to help us remove confounding is the inverse probability of treatment weighting (IPTW) [17]. This method relies on building a logistic regression model to estimate the probability score of the exposure (anemia/non-anemia) observed for the study population and using the probability score as a weight in subsequent analyses. The predicted probability score was calculated by putting all our baseline characteristics into the logistic regression model. By applying these weights to the study population, we can create a pseudo-population in which baseline characteristics are balanced across the 2 groups. We constructed 4 Cox regression models to test the robustness of the relationship between anemia and outcomes. In model 1, no confounder was adjusted. In model 2, confounders with p value 0.05 between the 2 groups were adjusted (age, gender, eGFR, SBP, diastolic BP, FRS, baseline CKD, race, serum creatinine, cancer history, cholesterol, blood glucose, and high-density lipoprotein). In model 3, we adjusted for the predicted probability score calculated by a logistic regression model. In model 4, the HR was calculated within the population after IPTW. A cumulative incidence plot for our outcomes was drawn both in pre-weighted and post-weighted populations with a log-rank test indicating the differences between the 2 groups. The follow-up time was censored at the end of the trial, death, loss to follow-up, or the occurrence of the outcomes being studied. The proportional hazards assumption was verified by checking the plot of Schoenfeld residuals.

Subgroup analyses were conducted to test the interaction effect (anemia status* subgroup) for our primary outcome among the following groups: age ($\ge$75/75 years), gender (male/female), FRS ($\ge$15%/15%), race (black/non-black), number of antihypertensive drugs at baseline ($\ge$2/2), previous CVD (Yes/No), medical history of cancer (Yes/No), SBP tertile at randomization (132 mmHg; 132 to 145 mmHg; $>$145 mmHg), and baseline CKD (eGFR $\ge$60/60 mL/min/1.73 m${}^2$). As for the R packages we used in our study, tableone, survey, and reshape2 were used for the inverse probability of treatment weighting; survival and survminer were used for Cox proportional hazards models; forestplot was used to draw forest-plot.

First, our hypothesis in this study is that anemia could be a risk factor for composite cardiovascular events and all-cause death among participants who received intensive BP treatment in SPRINT. To reinforce our findings, we also conducted the same analysis mentioned above among participants who received standard BP treatment. From our point of view, the relationship between anemia and outcomes found in the intensive BP treatment group may not exist in the standard BP treatment group. Second, we separated the whole SPRINT participants into 2 groups (anemia/non-anemia). The beneficial effect of intensive BP treatment compared with standard BP treatment in reducing cardiovascular events may not be identical among anemia and non-anemia participants. In other words, intensive BP treatment may not be suitable for participants with anemia.

The study flowchart (Fig. 1) shows the enrollment process of our study. There were 9361 participants enrolled in SPRINT, with 4678 participants randomized to the intensive BP treatment group and 4683 participants randomized to the standard BP treatment group. For the purpose of this study, we excluded participants in the standard BP treatment group. We further excluded 9 participants with missing information about anemia, 4 participants with missing information about cancer history, 4 participants with missing information about smoking status, 7 participants with missing information about aspirin usage, 9 participants with missing value of eGFR, 6 participants with missing value about serum creatinine, 202 participants with missing value about urine albumin-to-creatinine ratio, 20 participants with missing value of body mass index (BMI), and 23 participants with missing information about statin usage. Finally, there were 4394 participants included in this study. The dynamic change of BP after randomization among participants with anemia and non-anemia was shown in Supplementary Fig. 1.

Fig. 1.

Study flowchart. SPRINT, Systolic Blood Pressure Intervention Trial; BMI, body mass index; eGFR, estimated glomerular filtration rates.

Table 1 shows the baseline characteristics between the 2 groups before and after IPTW. Participants with anemia were older (mean age 68.86 versus 67.75, p = 0.01), more likely to be female (64.8% versus 31.8%, p 0.001), mostly Black race (37.2% versus 30.4%, p = 0.002), with lower FRS $\ge$15% (49.0% versus 63.4%, p 0.001), a higher percentage of cancer history (17.1% versus 12.1%, p = 0.001), higher prevalence of baseline CKD (41.5% versus 27.1%, p 0.001), higher SBP level at baseline (mean 140.96 mmHg versus 139.49, p = 0.043), lower DBP level at baseline (mean 76.96 versus 78.41, p = 0.008), lower eGFR (mean 66.15 versus 72.43, p 0.001), higher serum creatinine (mean 1.12 versus 1.07, p = 0.001), higher cholesterol (mean 193.75 versus 189.92, p = 0.046), lower blood glucose (mean 97.01 versus 99.18, p = 0.001), and more elevated high-density lipoprotein (mean 55.78 versus 52.46, p 0.001). After IPTW, baseline characteristics were similar between the 2 groups (Table 1, Supplementary Fig. 2).

Table 2 shows us the HRs for the association between anemia and outcomes. The presence of anemia was strongly associated with composite cardiovascular events after adjusting for potential confounders (HR 1.66, 95% CI 1.18–2.34, p = 0.004). The association remained statistically significant when we adjusted for propensity score (HR 1.61, 95% CI 1.14–2.27, p = 0.007) and did not change much even in the population after IPTW (HR 1.55, 95% CI 1.06–2.27, p = 0.024). The cumulative incidence plot indicates anemia was associated with the development of composite cardiovascular events, though the difference became marginally significant (p = 0.05) in the population after IPTW (Fig. 2A,B). As for the secondary outcome, participants with anemia had a higher rate of all-cause death compared with those without anemia (Table 2). The HR of all-cause death for participants with anemia compared with those without anemia was 1.75 (95% CI 1.15–2.66, p = 0.009) when adjusted for potential confounders, 1.67 (95% CI 1.09–2.54, p = 0.018) when adjusted for propensity score, and 1.61 (95% CI 1.00–2.57, p = 0.049) in the population after IPTW. The cumulative incidence plot indicates anemia was only associated with incident all-cause death in the pre-matched population (p = 0.001) but not in the population after IPTW (p = 0.1, Fig. 2C,D).

Fig. 2.

Cumulative incidence plot of time to outcome events by anemia status. (A) Impact of anemia on composite cardiovascular events in the pre-matched population. (B) Impact of anemia on composite cardiovascular events in post-matched population. (C) Impact of anemia on all-cause death in the pre-matched population. (D) Impact of anemia on all-cause death in the post-matched population.

Fig. 3 shows the interaction effect between anemia and prespecified groups on our primary outcome in the population after IPTW. Overall, no significant interaction effect was observed among all subgroups except for the subgroup of prior CVD (p for interaction = 0.04). Anemia seems to be a risk factor for composite cardiovascular events only among participants without prior CVD (HR 2.05, 95% CI 1.3–3.23) but not among participants with prior CVD (HR 0.86, 95% CI 0.43–1.73).

Fig. 3.

Interaction effect between anemia and prespecified subgroups on composite cardiovascular events in the post-matched population. HR, hazard ratio; FRS, Framingham risk score; CKD, chronic kidney disease; CVD, cardiovascular diseases.

First, anemia was not associated with composite cardiovascular events and all-cause death among participants treated with standard BP control (Supplementary Table 1). The HR of composite cardiovascular events and all-cause death for participants with anemia compared with those without anemia was 1.23 (95% CI 0.88–1.72, p = 0.221) and 1.19 (95% CI 0.78–1.82, p = 0.41) when adjusted for potential confounders, 1.15 (95% 0.83–1.62, p = 0.4) and 1.05 (95% CI 0.68–1.60, p = 0.84) when adjusted for propensity score, and 1.03 (95% 0.69–1.54, p = 0.89) and 1.08 (95% CI 0.66–1.78, p = 0.76) in the population after IPTW. Second, we divided the whole SPRINT participants into 2 groups (anemia and non-anemia) and investigated the interaction effect of anemia on the cardiovascular benefits of intensive BP control (Supplementary Table 2). The HR of composite cardiovascular events for participants who received intensive BP treatment compared with those who received standard BP treatment was 1.06 (95% CI 0.70–1.61, p = 0.77) for participants with anemia and 0.71 (95% CI 0.59–0.85, p 0.001) for participants without anemia (interaction p = 0.08). The HR of all-cause death for participants who received intensive BP treatment compared with those who received standard BP treatment was 1.17 (95% 0.70–1.95, p = 0.56) for participants with anemia and 0.67 (95% CI 0.53,0.84, p 0.001) for participants without anemia (interaction p = 0.05).

It is important to consider several limitations when interpreting the results of our study. First, this is a post-hoc analysis of SPRINT, and baseline characteristics between participants with anemia and non-anemia were unbalanced. Although we can balance baseline characteristics through IPTW, there may be other confounders that were not measured in SPRINT. Second, the anemia status was self-reported. The duration and degree of anemia can’t be adjusted in our model, so this may influence our observation. Prior treatment of anemia may also bias our observation. However, this kind of information was not provided in SPRINT. Third, as mentioned above, the causes of anemia are diverse. The association between anemia and outcomes may come from those undying diseases. Fourth, hemoglobin level was not measured in SPRINT. We were not able to analyze the impact of dynamic change in hemoglobin on our outcomes. Fifth, as shown in Supplementary Fig. 1, the BP of participants in the 2 treatment groups had a significant drop in the first 6 months after randomization. The dramatic change may increase the risk of cardiovascular disease, which was not considered in our study when investigating the impact of anemia. The landmark dynamic prediction model may help us gain insight into the dynamic changing effect of BP [34].