Serum Levels of Phosphorus, Parathyroid Hormone, and Calcium and Risks of Death and Cardiovascular Disease in Individuals With Chronic Kidney Disease: Title and subTitle BreakA Systematic Review and Meta-analysis FREE

Clinical practice guidelines are a powerful influence on management strategies for common conditions. When evidence that treating risk factors reduces disease and improves health outcomes is provided by large and well-conducted randomized controlled trials, guidelines can appropriately summarize the relevant data and widely disseminate recommendations for best practice. However, when practice guidelines promote therapeutic strategies without sufficient evidence of effectiveness or harms, overtreatment and widespread inappropriate use of medications, services, or devices may occur. Accordingly, guidelines may recommend health interventions that do not actually improve population outcomes or the quality of care, unnecessarily increase health care expenditure, and may even harm individuals who might be expected to receive small or negligible treatment benefit.

An important example is anemia in chronic kidney disease. Guidelines^{1 - 2} promoted anemia as a risk factor for mortality based on observational evidence of an association between low hemoglobin levels and the risks of death and hospitalization.^{3 - 4} It was argued that anemia required treatment in individuals with chronic kidney disease, but randomized controlled trials testing epoetins (an intervention that increases hemoglobin) showed that targeting higher hemoglobin levels caused harm,^{5 - 7} indicating that existing guidelines (based on observational data) were at best misleading.

A similar pattern has emerged in guidelines relating to the management of bone and mineral disorders in individuals with chronic kidney disease. Chronic kidney disease causes complicated changes in calcium and phosphorus metabolism⁸ for which numerous drugs including phosphorus binders, vitamin D compounds, and calcimimetic agents have been specifically developed and promoted. Nephrology guidelines recommend targets and treatment strategies to correct serum levels of phosphorus, calcium, and parathyroid hormone because observational data suggest there is an association between these potential risk biomarkers and vascular disease and death.^{9 - 10} However, to date, randomized controlled trials have not shown that treating mineral levels with existing treatment options reduces cardiovascular events or mortality.^{11 - 13} Nearly all such trials have been underpowered for mortality and cardiovascular outcomes, and a large comparative trial of phosphorus binders in hemodialysis found no evidence for benefit on mortality.¹⁴ Existing guidelines (based primarily on observational studies) may be inappropriately promoting treatment for abnormal mineral metabolism in individuals with chronic kidney disease.

Given the burden of cardiovascular morbidity in individuals with chronic kidney disease and the potential for serum levels of phosphorus, parathyroid hormone, and calcium to act as important modifiable risk factors for disease, we critically evaluated the evidentiary basis for an association between serum mineral levels and cardiovascular disease and mortality in individuals with chronic kidney disease. In particular, we sought to disaggregate those studies with more valid estimates of the association between mineral levels and outcomes through adjustment for known confounding variables.

We performed a systematic review and meta-analysis of cohort studies measuring the association between serum levels of phosphorus, parathyroid hormone, or calcium and death or cardiovascular events in individuals with chronic kidney disease according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines.¹⁵ A specialized librarian conducted a systematic and comprehensive database search of MEDLINE (1948 to December 2010) and EMBASE (1947 to December 2010) using a strategy that combined text word and Medical Subject Headings without language restriction (eTable 1). We also hand-searched the reference lists of primary studies, review articles, and clinical guidelines. We inspected the full text of any citation that appeared relevant.

We extracted data on the characteristics of study design, participants, exposures, and covariates together with the outcomes of all-cause mortality, cardiovascular mortality, and nonfatal cardiovascular events. Whenever available, we extracted the number of individuals at risk, and the adjusted hazard ratio or relative risk (RR) per unit baseline serum concentrations of phosphorus (1 mg/dL), parathyroid hormone (100 pg/mL), or calcium (1 mg/dL). The standard error of the estimate was also extracted or estimated from the reported 95% confidence interval (CI) if necessary. In addition, we extracted the adjusted hazard ratio or RR for each reported category of baseline serum concentrations of phosphorus, parathyroid hormone, or calcium. For each category of serum level exposure, we assigned the numerical value for the risk category as the midpoint of the upper and lower bound of the reported category.¹⁶ Whenever the category was open ended at the upper bound (eg, serum phosphorus >11 mg/dL), the assigned value recorded was the lower bound value of the category (11 mg/dL). Risk estimates for discrete patient cohorts reported within the same study (ie, participants receiving hemodialysis or peritoneal dialysis) were extracted separately.^{17 - 18} Whenever more than 1 publication reported data, we used all available reports to inform our data extraction while taking care to avoid duplication of information. Risks of bias in each study were assessed according to previously published methods.¹⁹

We plotted the reported log RR estimate for the outcomes of all-cause mortality, cardiovascular mortality, or nonfatal cardiovascular events against the corresponding category of serum level of phosphorus, parathyroid hormone, and calcium in each available cohort to qualitatively assess the linearity of the associations on the log scale. We then used the study-specific estimates of the RR per unit increase in serum level to calculate the summary estimates of RR (with 95% CI) per unit of increase in serum level of phosphorus, parathyroid hormone, or calcium. When this estimate was not available for a study, we used the methods described by Hartemink et al²⁰ to calculate RR estimates per unit of exposure based on the category-specific estimates of risk if the required data were reported. This method assumes that the relationship between the RR estimate per unit of exposure (level of serum phosphorus, parathyroid hormone, or calcium) is approximately linear.

Using studies where a linear association between log RR and exposure was reported or was estimable, we fitted a random-effects meta-regression model using the log RR per unit of exposure as the outcome to obtain a summary estimate of the log RR. The incremental units assigned were serum levels of phosphorus per 1-mg/dL increase relative to a referent value of 3.5 mg/dL (to convert to mmol/L, multiply by 0.323), parathyroid hormone per 100-pg/mL increase relative to a referent value of 200 pg/mL (to convert to ng/L, multiply by 1), and calcium per 1-mg/dL increase relative to a referent serum value of 8.4 mg/dL (to convert to mmol/L, multiply by 0.25). The meta-regression took account of the standard error for the estimate for each study. The summary log RR was exponentiated to obtain the summary estimate of RR.

We used subgroup analysis and meta-regression to assess the effect of adjustment for the key covariates of age, cardiovascular disease, diabetes status, time receiving dialysis (or estimated glomerular filtration rate for individuals not requiring dialysis), and race or ethnicity. The quality of adjustment was defined a priori as adequate when outcomes were adjusted for all 5 covariates and as partial when adjusted for 4 or fewer covariates. We also assessed the RR of mortality per unit of exposure for each of the following study-level covariates (when possible): age, proportion of men, race, stage of chronic kidney disease (dialysis, transplant, chronic kidney disease), dialysis adequacy, date of publication, duration of study, exposure measurement, and number of participants.

We conducted sensitivity analyses to evaluate whether using adjustment for cardiovascular disease as a criterion for assessing study quality altered our findings. Also, several studies were derived from the same data sources (eg, US Renal Data System, Dialysis Outcomes Practice Patterns Study). Whenever possible, we conducted sensitivity analyses by including studies from the same data source separately in summary analyses. For all analyses, a 2-tailed P value of less than .05 indicates statistical significance. Analyses were undertaken using SAS software version 9.2 (SAS Institute Inc, Cary, North Carolina) and Stata software version 11 (StataCorp, College Station, Texas) for the meta-regression.

A systematic search yielded 47 eligible studies (N = 327 644) in 49 cohorts of adults with chronic kidney disease (Figure 1; eTable 2). The population sizes of the studies, all published between 1995 and 2010, varied between 99 and 78 420 individuals. The 12 studies that enrolled more than 5000 individuals were derived from registry databases.^{21 - 32} Study duration ranged between 12 and 360 months. The vast majority of data were derived from cohorts requiring dialysis. Twenty-four studies (51%) assessed outcomes in 310 273 individuals receiving hemodialysis,^{21 - 44} and 3 studies assessed outcomes in 1124 participants receiving peritoneal dialysis.^{45 - 47} Five studies combined analysis in 3233 participants receiving hemodialysis and peritoneal dialysis,^{48 - 52} and 2 studies (n = 2251) analyzed outcomes for separate cohorts receiving hemodialysis and peritoneal dialysis.^{17 - 18} One study combined 98 individuals receiving dialysis, transplant recipients, and those with chronic kidney disease.⁵³ Four studies evaluated 1675 kidney transplant recipients.^{54 - 57} Eight studies (n = 8990) included individuals with chronic kidney disease in stages 3 through 5.^{58 - 65}

Outcomes were evaluated based on serum exposures at commencement of dialysis in 8 cohorts,^{18 ,38 ,41 ,43 ,48 - 51} at the time of kidney transplantation in 3 cohorts,^{55 - 56 ,62} and at the onset of chronic kidney disease in 2 cohorts.^{63 - 64} The definitions and sources of data for all-cause mortality, cardiovascular mortality, and cardiovascular events that were reported in the included studies appear in eTable 3. Cardiovascular mortality and cardiovascular events were generally defined as death or hospitalization due to myocardial infarction, atherosclerotic heart disease, cardiac arrest, or arrhythmia.

By current standards,¹⁹ we found that reporting of study methods was incomplete, suggesting risks of bias existed in the included studies (eFigure 1). Notably, data reporting completeness to follow-up (study attrition) and prognostic factor measurement were incomplete or unclear in approximately 90% of studies. Sixteen studies (34%) reported on methods used to adjudicate mortality events and, of these, 1 study reported formal adjudication procedures,³⁴ suggesting misclassification bias was possible in the remainder. Age was adjusted for in 46 studies (98%) and diabetes mellitus (81%), time receiving dialysis or duration of chronic kidney disease (66%), cardiovascular disease (47%), and race or ethnicity (47%). Thirteen studies (28%) reported risks of mortality or cardiovascular events according to the mean of serum levels.^{22 ,24 ,26 ,33 ,35 ,37 ,40 ,42 - 43 ,48 - 49 ,58 ,60} From these studies, we extracted risks for outcomes based on the mean of serum levels in the period before study commencement in 6 studies,^{22 ,26 ,35 ,40 ,49 ,58} mean of serum levels during the entire follow-up in 6 studies,^{24 ,33 ,37 ,43 ,48 ,56} and the mean of serum values in the period immediately before death or study completion in 1 study.⁴²

Of the 47 studies overall, graphical evaluation of the relationship between categories of serum exposure and risk was not possible in 7 studies (n = 74 114) because risk estimates for levels of serum exposure and an estimate of variance were not reported in an extractable format or a linear estimate of risk per unit increase in exposure was not extractable.^{23,24,43,47,56,62,63} Overall, 33 studies (n = 217 974) were not included in the meta-regression because they did not report the relationship between log RR per unit exposure, or this was not estimable from category-specific estimates.^{17,18,21-24,27-31,34-37,41,43-49,51,53-56,59,62-64}

On visual inspection, we found that the associations between serum levels of phosphorus, calcium, and parathyroid hormone and risks of all-cause mortality and other vascular outcomes were more consistent in studies with adequate adjustment for confounding variables than those studies with partial adjustment (Figure 2; eFigure 2 and eFigure 3). It should be noted that in general within most of the included studies, reported RRs derived from a base multivariate model that usually included age, race, sex, diabetes, and kidney function or duration of dialysis were not substantially altered by adjustment for additional potential confounding variables. Summary estimates for risks of all-cause mortality, cardiovascular mortality, and nonfatal cardiovascular events are shown in Figure 3.

In 12 partially adjusted studies (13 cohorts), the risk of death was higher with increasing levels of serum phosphorus (Figure 2).^{18 ,22 ,25 - 26 ,39 ,42 ,45 ,51 - 52 ,54 ,57 ,61} In 10 adequately adjusted studies, the relationship between serum phosphorus and all-cause mortality was more consistent and an increasing risk of death was apparent at higher levels of serum phosphorus (>5.5 mg/dL). Based on studies reporting a continuous RR for each 1-mg/dL increase in serum phosphorus^{25,26,33,38,40,42,50,52,58,60,61,65} together with a study in which the RR was estimated from reported event data,⁵⁷ we found that for every 1-mg/dL increase in serum phosphorus, the risk of mortality increased by 35% (RR, 1.35; 95% CI, 1.16-1.57) in the 3 adequately adjusted studies^58,60,65 and by 18% (RR, 1.18; 95% CI, 1.12-1.25) in the 13 available studies overall. We could not include the study by Young et al³² in these analyses because there was strong indication of nonlinearity based on reported risks for categories of serum phosphorus (Figure 3). However, addition of this study using the reported RR per unit increase in serum phosphorus made a negligible difference to the summary estimates. Only a single study with adequate adjustment for confounders reported the association between serum level of phosphorus and risk of cardiovascular mortality³² ; however, in the 3 studies reporting this outcome overall, the risk of cardiovascular mortality increased by 10% per 1-mg/dL increase in serum phosphorus (RR, 1.10; 95% CI, 1.06-1.13).^25,32,61 No data were available for the association between serum phosphorus and nonfatal cardiovascular events.

Whereas on visual inspection, 17 partially adjusted studies (19 cohorts) tended to show an increased risk of all-cause mortality at lower serum levels of parathyroid hormone (Figure 2),^{17 - 18 ,22 ,25 ,33 - 34 ,36 - 37 ,39 ,41 ,46 ,52 - 53 ,59 ,64} a summary analysis of the 4 available studies^{25 - 26 ,32 ,50} found no relationship between parathyroid hormone and all-cause mortality (Figure 3). Only a single adequately adjusted study reported the association between parathyroid hormone and cardiovascular mortality³¹ and no data were available for nonfatal cardiovascular events³⁰ in a format extractable for analysis; therefore, summary estimates for these risks were not calculated. However, when we combined the adequately and partially adjusted studies, we found no evidence for association between parathyroid hormone and cardiovascular mortality (Figure 3).

No association between all-cause mortality and serum calcium levels was observed in 2 adequately adjusted studies (Figure 3).^{30 - 32 ,44 ,49} Only a single adequately adjusted study provided data for the association between serum calcium levels and risks of cardiovascular mortality³² and no data were available for nonfatal cardiovascular events.³⁰ However, in all studies combined, there was evidence of association between serum calcium level and cardiovascular death^{18 ,31} (RR, 1.15; 95% CI, 1.08-1.23) (Figure 3).^{30 ,48}

We performed univariate meta-regression to identify other factors that might predict a relationship between serum level of phosphorus (eFigure 4) or calcium (eFigure 5) and risks of all-cause mortality. Univariate meta-regression was not possible for serum level of parathyroid hormone because fewer than 5 studies were available for the analyses. Risks of all-cause mortality associated with serum phosphorus and calcium were similar across different stages of chronic kidney disease. In the 4 studies^58,60,61,65 reporting outcomes for individuals with chronic kidney disease not yet requiring dialysis, the risk of all-cause mortality for each 1-mg/dL increase in serum level of phosphorus (RR, 1.29; 95% CI, 1.12-1.48) was similar to that observed in 8 studies^{25,26,33,38,40,42,50,52} of individuals requiring dialysis (RR, 1.17 [95% CI, 1.08-1.25]; P = .22). No evidence of an association between serum calcium and all-cause mortality was found in either individuals with earlier stages of chronic kidney disease (RR, 1.02; 95% CI, 0.81-1.29)^{32 ,58} or those requiring dialysis (RR, 1.09 [95% CI, 1.00-1.18]; P = .63).^{25,26,32,42,50,52,57}

We assessed the robustness of our assumption that adjusting for cardiovascular disease was necessary to define a study as adequately adjusted (in addition to age, race, duration of dialysis or kidney function, and presence of diabetes mellitus). When excluding cardiovascular disease as a criterion for adjustment quality, 6 studies (7 cohorts) were adjudicated as adequately adjusted.^{17 ,22 ,24 ,26 ,52 ,60} Of these, 2 studies^{26 ,52} contributed data to the summary analysis for phosphorus and all-cause mortality. Categorizing these studies as adequately adjusted in the meta-analysis for this outcome did not alter the association we found between serum level of phosphorus and death (RR, 1.24; 95% CI, 1.12-1.37). These 2 studies were also in the meta-analysis for calcium and all-cause mortality. Adjudication of these studies as adequate did not meaningfully alter the summary estimate we found for the association between serum level of calcium and all-cause mortality (RR, 1.09; 95% CI, 0.98-1.21).

Several studies were derived from the same source; 4 studies were derived from the US Renal Data System^{21 ,23 ,28 ,30} and 5 studies included individuals sourced from the Dialysis Outcomes and Practice Patterns Study.^{25 ,27 ,29 ,31 - 32} Of all the meta-analyses, 2 studies from the same data source contributed to the summary estimates for parathyroid hormone and calcium and all-cause mortality. Entering these 2 studies^{25 ,32} separately into the meta-analyses for all studies combined and only for adequately adjusted studies did not meaningfully change the associations we observed between serum parathyroid hormone and serum calcium with all-cause mortality.

When receiving treatment designed to reach proposed serum target levels for phosphorus, parathyroid hormone, and calcium, individuals with chronic kidney disease, clinicians, and policy makers need to know whether serum levels have any evidence of association with health outcomes. Prior treatment trials for mineral and bone disorder in chronic kidney disease found no evidence for clinical benefit for any of the agents tested so far.^{11 - 13} Our study provides the first systematic and critical review using data from cohort studies of the association between serum levels of phosphorus, parathyroid hormone, and calcium and death and cardiovascular outcomes in individuals with chronic kidney disease.

Overall, we were unable to demonstrate any strong or consistent association between death and serum levels of parathyroid hormone and calcium in individuals with chronic kidney disease. While there may be an association between death and higher serum levels of phosphorus irrespective of the stage of kidney disease, high-quality data for an association between serum levels of phosphorus, parathyroid hormone, and calcium with cardiovascular mortality were available only in single cohort studies and summary estimates for cardiovascular events were not available. Consequently, based on the available cohort data and the absence of randomized controlled trials, the evidentiary basis for current clinical guideline–recommended targets of serum phosphorus, parathyroid hormone, and calcium in chronic kidney disease is poor.^{9 - 10}

Broad adoption of health care practices that have insufficient evidence for safety or efficacy (in this case targeting serum mineral levels in individuals with chronic kidney disease) may lead to considerable unintended harm. Examples in medicine include the widespread adoption of treatment using encainide and flecainide to suppress ventricular tachyarrhythmias after myocardial ischemia and hormone therapy in postmenopausal women to prevent cardiovascular disease until randomized trials unequivocally demonstrated such interventions were harmful.^{66 - 67}

We now demonstrate that the evidence is similarly lacking for serum mineral target levels as risk factors for poor health outcomes in individuals with chronic kidney disease despite nearly universal adoption of such targets in the field of nephrology.^{68 - 69} The current data do not support the hypothesis that individuals with chronic kidney disease should have treatment to achieve targeted levels of serum parathyroid hormone or calcium to reduce mortality or cardiovascular morbidity, except at extreme levels in which hypocalcemia and hypercalcemia result in immediate, clinically apparent adverse events such as tetany and seizures. Furthermore, treating high phosphorus levels is linked to a substantial pill burden that is associated with lower quality of life in individuals with chronic kidney disease.⁷⁰ While we do not conclude that normalizing serum levels of calcium or phosphorus or avoiding upper or lower extremes of serum level of parathyroid hormone is futile, high-quality evidence is required before specific treatment should be advocated strongly.

Accumulating cohort data demonstrate that associations exist between vitamin D and phosphorus binder treatment and lower mortality and cardiovascular events in individuals with chronic kidney disease that may be independent of their treatment effects on serum levels of phosphorus and parathyroid hormone.^{71 - 74} Vitamin D receptor activation may attenuate left ventricle structural abnormalities, reduce left ventricle mass, and improve diastolic function.⁷⁵ Given the associations between phosphorus level and cardiovascular events we found in the current study together with these uncontrolled data suggesting benefit for specific interventions in chronic kidney disease, we strongly advocate that large simple placebo-controlled randomized trials be conducted of vitamin D, calcimimetic agents (allosteric modulators of parathyroid calcium–sensing receptors that lower parathyroid hormone secretion), or phosphorus binders in individuals with chronic kidney disease.

The earlier focus by the nephrology community on hemoglobin targets is a relevant cautionary tale. Because hemoglobin targets became the research priority, there was considerable delay in commencing a large placebo-controlled trial for anemia in individuals with chronic kidney disease. Similarly, the limited evidentiary basis for an association between serum mineral levels and health outcomes in the present analysis suggests that target trials of serum levels of phosphorus, parathyroid hormone, and calcium should not be promoted in individuals with chronic kidney disease.⁷⁶

Our study has limitations that should be considered. First, our conclusions are supported by low-quality data because summary effects are derived from uncontrolled cohort studies that are vulnerable to the unpredictable confounding effects of measured and unmeasured variables. Notably, within most of the included studies, RRs from a basic adjusted model were not altered by further adjustment for additional variables, indicating that partial adjustment was not an important limitation of these studies.

Second, we have assumed that the relationship between serum levels of phosphorus, parathyroid hormone, and calcium and health outcomes is approximately linear. While we assessed the validity of this assumption in individual studies using a regression analysis, the small number of exposure categories within most of the studies meant that we had limited power to detect nonlinear relationships between serum levels and log risks of mortality. Third, we have not assessed study data for the association between serum levels of alkaline phosphatase or vitamin D levels and mortality, which are both emerging as predictors of outcomes in populations with and without chronic kidney disease in cohort studies^{77 - 79} and are potential effect modifiers of the relationships between serum targets and outcomes that we have observed. We also have not studied the effects of serum mineral levels on other relevant clinical outcomes such as musculoskeletal pain or itch. Finally, phosphorus and calcium may function poorly as biomarkers because they represent only a fraction of body stores and have substantial intraindividual variability over time (despite assay precision).

In conclusion, the evidence for an association between serum levels of phosphorus, calcium, or parathyroid hormone as risk factors for outcomes in individuals with chronic kidney disease (at any stage) is currently insufficient to inform clinical decision making, policy, or practice guidelines. Large placebo-controlled randomized trials of vitamin D compounds, phosphorus binders, and calcimimetic agents are now needed to evaluate whether treating mineral disorders improves health outcomes for individuals with chronic kidney disease.