Abstract
Rationale & Objective:
Optimal approaches to treat secondary hyperparathyroidism (SHPT) in patients on maintenance hemodialysis (HD) are not established in randomized controlled trials (RCTs).
Study Design:
Two observational clinical trial emulations.
Setting & Participants:
Both emulations included adults receiving in-center HD from a national dialysis organization. Included patients had SHPT in the period between 2009–2014, were insured for ≥180 days by Medicare as primary payer, and did not have contraindications or poor health status limiting theoretical trial participation.
Exposures:
The PTH Target Trial Emulation included patients with new-onset SHPT (first PTH 300–600 pg/ml), with 2 arms defined as up-titration of either vitamin D sterols or cinacalcet within 30 days (lower target) or no up-titration (higher target). The Agent Trial Emulation included patients with a PTH ≥300 pg/ml while on ≥6 mcg weekly of vitamin D sterol (paricalcitol equivalent dose) and no prior history of cinacalcet. The 2 arms were defined by the first dose or agent change within 30 days [vitamin D-favoring (vitamin-D was up-titrated) vs. cinacalcet-favoring (cinacalcet was added) vs. non-defined (neither applies)]. Multiple trials per patient were allowed in Trial 2.
Outcomes:
The primary outcome was all-cause death over 24 months; secondary outcomes included cardiovascular (CV) hospitalization or the composite of CV hospitalization or death.
Analytical Approach:
Pooled logistic regression.
Results:
There were 1,152 patients in the PTH Target Trial (635 lower target, 517 higher target). There were 2,726 unique patients with 6,727 patient-trials in the Agent Trial (6,268 vitamin D-favoring trials and 459 cinacalcet-favoring trials). The lower PTH target approach was associated with reduced adjusted hazard of death (HR 0.71; 95% CI 0.52, 0.93), CV hospitalization (HR 0.78; 95% CI 0.63, 0.98), and their composite (HR 0.74; 95% CI 0.61, 0.89). The cinacalcet-favoring approach demonstrated lower adjusted hazard of death compared to the vitamin D-favoring approach (HR 0.79; 95% CI 0.62, 0.99), but not of CV hospitalization or the composite outcome.
Limitations:
Potential for residual confounding; low use of cinacalcet with low power.
Conclusions:
SHPT management focused on lower PTH targets may lower mortality and CV disease in patients receiving HD. These findings should be confirmed in a pragmatic randomized trial.
Keywords: pharmacoepidemiology, mineral metabolism, calcimimetic, vitamin D, parathyroid hormone
Plain Language Summary
Optimal approaches to treat secondary hyperparathyroidism have not been established in randomized controlled trials. Data from a national dialysis organization was used to identify patients with secondary hyperparathyroidism (SHPT) in whom escalated treatment may be indicated. The approach to treatment was defined based on observed upward titration of SHPT-controlling medications: earlier titration (lower target) vs. delayed titration (higher target); and the choice of medication (cinacalcet vs. vitamin D sterols). In the first trial emulation, we estimated a 29% lower rate of death and 26% lower rate of cardiovascular disease or death for patients managed with a lower vs. higher target approach. Cinacalcet vs. vitamin D-favoring approaches were not consistently associated with outcomes in the second trial emulation. This observational study suggests the need for additional clinical trials of SHPT treatment intensity.
Introduction
Secondary hyperparathyroidism is a common problem in patients with kidney failure treated with hemodialysis. Vitamin D sterols, including calcitriol or vitamin D analogs, and calcimimetic agents are available to treat this disorder and were adopted based on data that they can safely and effectively lower parathyroid hormone (PTH) in randomized controlled trials. 1 In the pre-treatment era, secondary hyperparathyroidism was highly morbid contributing to bony pain, deformity, fractures, and diffuse vascular calcification.2 In the last decade, secondary hyperparathyroidism and related abnormalities in the bone and mineral axis, referred to collectively as chronic kidney disease mineral and bone disorder (CKD-MBD), have been widely implicated in the pathogenesis of cardiovascular disease, the leading cause of death in patients with kidney failure.3, 4 Whether more intensive therapy of secondary hyperparathyroidism targeting a lower PTH level, or utilizing different preferred agents can reduce cardiovascular outcomes and mortality in patients maintained on hemodialysis remains an unsettled question in the field.
The lack of trial evidence guiding PTH target levels and ideal agents is reflected in current CKD-MBD treatment guidelines and in equipoise observed among providers. For instance, the latest 2017 CKD-MBD guidelines released by the Kidney Disease: Improving Global Outcomes workgroup suggests “maintaining intact PTH levels in the range of approximately 2 to 9 times the upper normal limit for the assay”, or approximately 150–600 pg/ml.5 This range was expanded from 150–300 pg/ml that was initially recommended by the National Kidney Foundation’s Kidney Disease Outcome Quality Initiative in 2003.6 There is also limited guidance on first-line agents suggesting that providers may use “calcimimetics, calcitriol, or vitamin D analogs, or a combination of calcimimetics with calcitriol or vitamin D analogs”.5 We have previously shown a high and growing level of variation in approaches to managing this problem across dialysis facilities, reflecting genuine and justified equipoise about best practices for patient care.7 Trial emulation is an increasingly recognized and cost-effective framework utilizing existing data to inform the design of future clinical trials particularly in areas of care, such as secondary hyperparathyroidism, where best practices are uncertain.8–10 We conducted two trial emulations using electronic health record data from a national dialysis provider, comparing different approaches to PTH targets and agents to plan definitive trials related to secondary hyperparathyroidism, mortality, and cardiovascular disease in patients treated with hemodialysis.
Materials and Methods
This study is an emulation of two pragmatic comparative effectiveness trials comparing alternative treatment approaches to secondary hyperparathyroidism in patients on maintenance hemodialysis (Figure 1). Detailed methods are described in Item S1.
Figure 1. Conceptual Diagram of Clinical Trial Emulation Framework.
Major criteria related to the population, intervention, comparator, and outcomes for the ideal trial and the emulated trial are contrasted according to the two clinical questions.
‘Screening’ Study Population and Measurements
Our study included adult patients treated at facilities affiliated with Dialysis Clinic, Inc. (DCI), a moderate-sized, not-for-profit national dialysis organization in the United States. Patients with incident kidney failure initiating dialysis at a DCI facility between January 1, 2006 and June 15, 2014 were considered to coincide with availability of cinacalcet, availability of Medicare Part D, and an adequate potential follow up duration in our dataset. Patients meeting the screening study population criteria were then prospectively evaluated for eligibility in our Trials based on laboratory and medication criteria. This study was approved by the Institutional Review Board at Duke University School of Medicine under Protocol #Pro00062730, and a waiver of informed consent was granted. USRDS files were authorized under Data Use Agreement #DUA 2016–41e.
The PTH Target Trial Emulation: Eligibility and Inclusion
Eligibility for the Target Trial required a PTH 300–600 pg/ml observed January 1, 2009 or later to coincide with updated practice guidelines on CKD-MBD in 2009.11 No prior PTH >600 pg/ml was allowed and the eligible PTH was the first in range since dialysis initiation. Patients had to be able to potentially contribute 750 days of follow-up (1, 30-day treatment observation window and 24, 30-day follow-up periods). To allow ascertainment of comorbidities using claims, we required that all patients have at least 180 days of continuous Medicare Primary Payer coverage prior to the eligibility date, but this could include the 90-day waiting period for coverage.
We excluded patients who may not be optimal candidates for our target trial because they were not on in-center hemodialysis at the time of eligibility, had contraindications to medication titration, or were not optimal candidates for a trial due to either poor health status limiting follow-up or poor dialysis treatment adherence. Details of exclusion criteria are described in Item S2.
The PTH Target Trial Emulation: Treatment Strategies and Assignment
We defined lower vs. higher PTH target approaches based on the medication titration response within 30 days. Medications were ascertained by billing codes and dose conversions provided in Item S3. Patient with at least one unambiguous upward titration of vitamin D sterol or cinacalcet were assigned to the lower target arm. Patients with downward titration, no titration of either class, or class switching were assigned to the higher target arm. We produced both intention-to-treat estimands and per-protocol estimands that censor patients if non-adherent. Definitions for adherence are described in Item S2.
The PTH Target Trial Emulation: Follow-up period
Immediately following eligibility was a 30-day treatment observation window. Follow-up for outcomes then occurred through 24 subsequent 30-day periods. Patients were censored at kidney transplant, loss of Medicare Primary Payer status, transfer from DCI facilities (per-protocol analyses only), or the end of follow-up.
The PTH Target Trial Emulation: Study Outcomes
The primary outcome was all-cause mortality. Secondary outcomes include first hospitalization with a cardiovascular cause, and a composite outcome of mortality or first hospitalization with a cardiovascular cause. Death was ascertained using the USRDS 2018 Core files. Hospitalization with a cardiovascular cause was ascertained based on the primary billed discharge diagnosis using USRDS Hospitalization files and administrative codes.12 Detailed outcome definitions can be found in Item S4.
The Agent Trial Emulation: Eligibility and Inclusion
Eligibility for the Agent Trial required a PTH ≥300 pg/ml. Because cinacalcet is rarely used first-line for secondary hyperparathyroidism in this era, all patients must have been treated with at least 6 mcg of paricalcitol equivalent weekly. To avoid bias from prior failed therapy, patients could not have a history of prior use of cinacalcet before their eligible date. Any timepoint that met these criteria (i.e., PTH ≥300 pg/ml; using ≥6 mcg of paricalcitol equivalent weekly; and with no history of cinacalcet use) could be used for the Agent Trial. This means that repeated patient-trials could be contributed by each patient. Additional exclusions were applied as described for the Target Trial. An expanded description is available in Item S1.
The Agent Trial Emulation: Treatment Strategies and Assignment, Follow up, and Outcomes
The cinacalcet-favoring approach was defined by initiation of cinacalcet within 30 days, whereas the vitamin D-favoring approach was defined by upward titration of a vitamin D sterol within 30 days. If no titration of PTH-controlling medication occurred in the 30 days following PTH ≥300 pg/ml, then prescribing behavior was inconsistent with either approach and the patient-trial was not included. Definitions of adherence are described in Item S2. Follow up and Study Outcomes were the same as for the Target Trial except that we allowed repeated eligibility.
Causal Contrast and Statistical Analysis for Both Trials
We had two causal contrasts of interest: intention-to-treat and the per-protocol effect.
Intention-to-treat Analysis
We estimated the hazard ratios for the intention-to-treat analysis using pooled logistic regression with a treatment arm indicator and a quadratic functional form for time (i.e., number of 30 day periods since start of follow-up).13 Potential confounders, measured at the date of eligibility were then added to the model including demographics, selected comorbidities, Liu comorbidity index,14 original cause of kidney failure, any hospitalization with cardiovascular cause in the prior 6 months, body mass index (BMI), PTH level, phosphorus level, calcium level (albumin-corrected), Kt/V, dialysate calcium, dialysis duration, and vascular access. Confidence intervals were constructed using cluster bootstrapping sampled at patient level for the Target Trial and patient-trial level for the Agent Trial with 500 samples.
Per-protocol and Secondary Analysis
For the per-protocol analysis, we censored patients if they deviated from the specified protocol for their original assigned treatment approach. We used stabilized inverse probability weighting (IPW) to adjust for potential selection bias induced by censoring on non-adherence.15, 16 Standardized survival curves were computed to aid in interpretation of hazard ratios and absolute risks as recommended in prior literature.17, 18 Detailed methods are provided in Item S1.
Sub-analyses and Sensitivity Analyses
We conducted several sub-analyses to further characterize our main intention-to-treat results. Recognizing that hazard can change over time, we calculated hazard ratios via pooled logistic regression for alternative endpoints of 6-, 9-, 12-, and 18-months.18 In addition, due to prior trial evidence that cinacalcet may be more efficacious in older adults,19 we tested for heterogeneity of treatment effects by age at eligibility (age <65 versus ≥65 years) by adding interaction terms with treatment.
We conducted two major sensitivity analyses to evaluate the robustness of our main analyses. First, we restricted our eligible cohort to only those patients incident to dialysis in 2009 or later, after the change in clinical practice guidelines. Second, we relaxed the requirement for continuous Medicare Primary Payer coverage prior to eligibility from 180 to 90 days. Finally, our criteria for the Agent Trial left open the possibility that the same titration could be linked to more than one PTH result, leading to distinct but closely overlapping trials. In a sensitivity analysis, we randomly selected one PTH result to associate with each qualifying titration.
Results
Between January 1, 2006 to June 15, 2014, 26,265 patients initiated dialysis at DCI, of which 21,949 had at least 1 PTH laboratory in 2009 or later.
The PTH Target Trial
Of the 21,949 patients who had at least one PTH result, ultimately 635 patients could be assigned to the lower target arm and 517 to the higher target arm. Major exclusions included lack of PTH in range after 2009 and lack of 180-day lookback period (Figure 2). Median age of patients in the Target Trial was 69.1 years, 45.0% were female, median PTH level was 353 pg/ml without difference (p=0.2) between treatment arms. The lower target arm had a higher proportion of Black patients, a higher proportion of individuals that were female and were more likely to have an earlier year of dialysis incidence and an earlier year of trial entry (Table 1).
Figure 2. Detailed Study Flow for Inclusion and Exclusion in the Target Trial: Lower vs. Higher Parathyroid Hormone (PTH) Target.
The flow diagram demonstrates inclusion of patients incident to dialysis at Dialysis Clinic, Inc (DCI) facilities in the study period. Exclusion are enumerated sequentially in boxes to right and left with the included participants at each step in the central column. Final included patients are reported for the primary study population overall and for the subpopulation incident to dialysis in 2009 or later, which represents a sensitivity analysis population.
Table 1.
Characteristics at Time of Eligibility by Treatment Group for the Target Trial Emulation
| Lower Target (N=635) |
Higher Target (N=517) |
Total (N=1152) |
p value | |
|---|---|---|---|---|
| Sociodemographics: Median (IQR) or n (%) | ||||
| Age (years) | 68.4 (59.3, 76.0) | 69.7 (60.6, 77.5) | 69.1 (59.8, 76.7) | 0.041 |
| Female sex | 304 (47.9%) | 214 (41.4%) | 518 (45.0%) | 0.032 |
| Race | <0.0012 | |||
| Black | 176 (27.7%) | 88 (17.0%) | 264 (22.9%) | |
| White | 429 (67.6%) | 388 (75.0%) | 817 (70.9%) | |
| Other | 30 (4.7%) | 41 (7.9%) | 71 (6.2%) | |
| Hispanic ethnicity | 29 (4.6%) | 44 (8.5%) | 73 (6.3%) | 0.0072 |
| History of smoking | 57 (9.1%) | 52 (10.1%) | 109 (9.6%) | 0.52 |
| Part D Enrollment | 0.22 | |||
| Enrolled: total copayment subsidy | 75 (11.8%) | 52 (10.1%) | 127 (11.0%) | |
| Enrolled: moderate copayment subsidy | 157 (24.7%) | 112 (21.7%) | 269 (23.4%) | |
| Enrolled: small or no copayment subsidy | 223 (35.1%) | 212 (41.0%) | 435 (37.8%) | |
| Not enrolled | 180 (28.3%) | 141 (27.3%) | 321 (27.9%) | |
| Health history | ||||
| Hypertension | 573 (91.2%) | 463 (90.3%) | 1036 (90.8%) | 0.62 |
| Diabetes mellitus | 475 (74.8%) | 373 (72.1%) | 848 (73.6%) | 0.32 |
| Cancer | 96 (15.1%) | 84 (16.2%) | 180 (15.6%) | 0.62 |
| Comorbidity Index | 9 (4, 11) | 8 (5, 11) | 8 (5, 11) | 0.91 |
| Cardiovascular hospitalizations past 6m | 69 (10.9%) | 55 (10.6%) | 124 (10.8%) | 0.92 |
| Functional Limitation | 113 (18.0%) | 91 (17.7%) | 204 (17.9%) | 0.92 |
| Recent clinical characteristics | ||||
| BMI (kg/m2) | 27.3 (23.8, 32.5) | 27.3 (23.3, 31.8) | 27.3 (23.7, 32.2) | 0.21 |
| PTH (pg/ml) | 354 (311, 414) | 353 (321, 402) | 353 (317, 410) | 0.21 |
| Phosphorus (mg/dl) | 5.1 (4.3, 5.9) | 5.2 (4.4, 6.3) | 5.1 (4.3, 6.1) | 0.041 |
| Albumin Corrected Calcium (mg/dl) | 9.1 (8.7, 9.4) | 9.0 (8.7, 9.4) | 9.1 (8.7, 9.4) | 0.31 |
| Dialysis characteristics | ||||
| Dialysis vintage (years) | 1.4 (0.9, 2.2) | 1.4 (0.9, 2.1) | 1.4 (0.9, 2.2) | 0.91 |
| Primary cause of kidney failure | 0.42 | |||
| Diabetes mellitus | 325 (51.2%) | 275 (53.2%) | 600 (52.1%) | |
| Hypertension | 164 (25.8%) | 120 (23.2%) | 284 (24.7%) | |
| Glomerulonephritis | 45 (7.1%) | 47 (9.1%) | 92 (8.0%) | |
| Other cause | 101 (15.9%) | 75 (14.5%) | 176 (15.3%) | |
| Mean dialysis time (minutes) | 209.9 (184.2, 226.7) | 210.7 (193.8, 232.1) | 210.0 (187.7, 229.1) | 0.021 |
| Dialysate calcium (mEq/l) | 0.32 | |||
| <2.5 (%) | 112 (17.6%) | 81 (15.7%) | 193 (16.8%) | |
| ≥2.5 (%) | 523 (82.4%) | 436 (84.3%) | 959 (83.2%) | |
| Kt/V | 1.5 (1.4, 1.7) | 1.5 (1.4, 1.7) | 1.5 (1.4, 1.7) | 0.41 |
| Primary vascular access | 0.22 | |||
| Fistula | 398 (62.7%) | 351 (67.9%) | 749 (65.0%) | |
| Graft | 108 (17.0%) | 71 (13.7%) | 179 (15.5%) | |
| Catheter | 129 (20.3%) | 95 (18.4%) | 224 (19.4%) | |
| Secular trends | ||||
| Year of incident kidney failure | <0.0012 | |||
| 2006 | 40 (6.3%) | 21 (4.1%) | 61 (5.3%) | |
| 2007 | 92 (14.5%) | 38 (7.4%) | 130 (11.3%) | |
| 2008 | 154 (24.3%) | 62 (12.0%) | 216 (18.8%) | |
| 2009 | 146 (23.0%) | 82 (15.9%) | 228 (19.8%) | |
| 2010 | 95 (15.0%) | 112 (21.7%) | 207 (18.0%) | |
| 2011 | 43 (6.8%) | 90 (17.4%) | 133 (11.5%) | |
| 2012 | 45 (7.1%) | 62 (12.0%) | 107 (9.3%) | |
| 2013 | 17 (2.7%) | 42 (8.1%) | 59 (5.1%) | |
| Year trial entry | <0.0012 | |||
| 2009 | 203 (32.0%) | 54 (10.4%) | 257 (22.3%) | |
| 2010 | 154 (24.3%) | 71 (13.7%) | 225 (19.5%) | |
| 2011 | 100 (15.7%) | 123 (23.8%) | 223 (19.4%) | |
| 2012 | 83 (13.1%) | 111 (21.5%) | 194 (16.8%) | |
| 2013 | 54 (8.5%) | 90 (17.4%) | 144 (12.5%) | |
| 2014 | 41 (6.5%) | 68 (13.2%) | 109 (9.5%) | |
Abbreviations: PTH=parathyroid hormone; BMI=body mass index
p-value calculated for two-sided Kruskal Wallis test with 5% significance
p-value calculated for two-sided Chi-square test with 5% significance
When examining mean and median PTH levels in the cohort over the course of follow-up, those in the lower target arm had lower PTH levels in the first few months, with levels approaching the higher target arm over time (Figure 3a). These trends represent unadjusted within-cohort and not necessarily within-patient effects, as some patients are censored or have events over follow-up, changing the composition of the cohort over time. An exploration of titration patterns after baseline demonstrated that the lower target arm continued to have higher rates of titration overall. Titrations included higher rates of upward titration and downward titrations, which may have been responses to over suppression of PTH or side effects (Figure 3b).
Figure 3. Levels of Parathyroid Hormone and Medication Titrations Over Time in Patients Participating in the Target Trial.
In Panel A, the unadjusted achieved parathyroid hormone (PTH) level in the cohort over 24 months on study is demonstrated. Patients following a lower target approach are depicted in solid links with higher target approach depicted in dashed lines. Bars represent 95% confidence intervals around each monthly mean level (pictured in black). For comparison, monthly median levels are provided for each group in red. Not all patients are represented in each summary distribution due to censoring over time with the number included presenting in the table below the X-axis. Missing PTH values in any given month are carried forward from the most recent measure. In Panel B, loess smoothed functions of proportion of patients with medication titrations post-baseline are depicted in each group over follow-up with their 95% confidence intervals. Solid lines depict an upward titration in PTH-controlling medications. Dashed lines depict a downward titration in PTH-controlling medications. Blue lines in each set reflect the lower target treatment group and red lines reflect the higher target treatment group. Time 0 is 30 days after the eligible PTH, when follow-up begins. ‘Lower Target’ exposure assignment was based on at least one unambiguous upward titration in a PTHcontrolling medication in the 30 day exposure assessment window that precedes this follow-up period.
Amongst the 1,152 individuals included in the Target Trial, there were 287 deaths (24.9%), 410 patients (35.6%) with at least one hospitalization for cardiovascular cause, and 560 composite outcome events representing 48.6% of trial participants. For the intention-to-treat analysis, the estimated hazard ratio for death was 0.71 (95% CI: 0.52 to 0.93) for the lower vs. the higher target arm (Table 2). The estimated hazard ratio of first hospitalization with cardiovascular cause was 0.78 (95% CI: 0.63 to 0.98), and the estimated hazard ratio for the composite outcome was 0.74 (95% CI: 0.61 to 0.89) comparing lower to higher target. Estimated two-year probability for event free survival for the composite outcome was 49.5% (95% CI: 39.5 to 61.1) for the lower target arm and 57.4% (95% CI: 52.9 to 61.9) for the higher target arm, consistent with primary analysis but not statistically different (risk difference −7.9%; 95% CI,−19.6 to 3.9; Figure S1). Overall, there were no statistical differences in the treatment effects by age (Table S1).
Table 2.
Pooled Logistic Estimation of Hazard Ratio for Death, First Cardiovascular Hospitalization, and a Composite over 24 Months for Lower Compared to Higher PTH Target Patients in the Target Trial
| Intention-to-Treat | Per-Protocol | ||
|---|---|---|---|
| Unadjusted | Adjusted | Adjusted | |
| Outcome | Hazard Ratio (95% CI) | Hazard Ratio (95% CI) | Hazard Ratio (95% CI) |
| Death | 0.74 (0.57, 0.94) | 0.71 (0.52, 0.93) | 0.80 (0.47, 1.26) |
| Cardiovascular Hospitalization | 0.81 (0.66, 0.99) | 0.78 (0.63, 0.98) | 0.79 (0.55, 1.24) |
| Composite Death and Cardiovascular Hospitalization | 0.78 (0.66, 0.92) | 0.74 (0.61, 0.89) | 0.78 (0.55, 1.12) |
Note: Hazard ratios reflect the lower target group compared to the higher target group
Prescription patterns over time indicated relatively high adherence to original assigned treatment arms with ~90% staying on protocol for the higher target arm and 70–80% staying on the lower target arm over follow-up (Figure S2). In the per-protocol analysis, there were 178 deaths (15.5%), 247 hospitalizations with cardiac cause (24.1%), 344 composite events (29.9%), with estimated hazard ratio of death was 0.80 (95% CI: 0.47 to 1.26) for lower vs. higher target. Additional per-protocol effects are presented in Table 2. When examining alternative timeframes for ITT and per-protocol hazard ratios (Table S2), results were similar, but per-protocol effects were notably less precise.
Results for the Target Trial were largely similar when the sample was limited to patients incident to dialysis in 2009 or later (Table S3) and when the 180-day look-back requirement was loosened to require only 90-days (Table S4).
The Agent Trial
Of the 21,949 patients that had PTH lab results in 2009 or later, 17,987 (82%) had at least one measured PTH result ≥ 300 pg/ml resulting in 181,096 potential patient-trials for the Agent Trial. Detailed exclusion criteria are presented in Figure 4 yielding a final sample size of 459 patient-trials for the cinacalcet-favoring approach and 6,268 patient-trials for the vitamin D-favoring approach.
Figure 4. Detailed study flow for inclusion and exclusion in the Agent Trial: Cinacalcetfavoring vs. Vitamin D-favoring Treatment of Parathyroid Hormone (PTH).
The flow diagram demonstrates inclusion of patients incident to dialysis at Dialysis Clinic, Inc (DCI) facilities in the study period. Exclusion are enumerated sequentially in boxes to right and left with the included participants at each step in the central column. Final included patients are reported for the primary study population overall and for the subpopulation incident to dialysis in 2009 or later, which represents a sensitivity analysis population. A small number of duplicate trials are eligible in the cinacalcet-favoring arm because more than on PTH value met the inclusion criteria within 30 days and before the trial-defining titration. These closely linked trials were removed from the cinacalcet-favoring and vitamin D-favoring arm in sensitivity analyses.
Median age across all patient-trials in the Agent Trial was 65.2 years, 46.4% were female, median PTH level was 447 pg/ml with a higher baseline PTH in the cinacalcet-favoring arm versus the vitamin D-favoring arm (Table 3). The cinacalcet-favoring arm had a lower median age, higher levels of Part D copayment subsidy, fewer females, and modestly lower median comorbidity scores, each compared to the vitamin D-favoring arm (Table 3).
Table 3.
Characteristics at Time of Eligibility by Treatment Group for Each Patient-Trial Included in the Agent Trial Emulation
| Cinacalcet-favoring (N=459) |
Vitamin D-favoring (N=6268) |
Total (N=6727) |
p-value | |
|---|---|---|---|---|
| Sociodemographics: Median (IQR) or n (%) | ||||
| Age (years) | 63.1 (53.0, 71.2) | 65.4 (56.1, 72.9) | 65.2 (55.9, 72.8) | <0.0011 |
| Female sex | 191 (41.6%) | 2932 (46.8%) | 3123 (46.4%) | 0.032 |
| Race | 0.52 | |||
| Black | 245 (53.4%) | 3381 (53.9%) | 3626 (53.9%) | |
| White | 191 (41.6%) | 2640 (42.1%) | 2831 (42.1%) | |
| Other | 23 (5.0%) | 247 (3.9%) | 270 (4.0%) | |
| Hispanic ethnicity | 23 (5.0%) | 394 (6.3%) | 417 (6.2%) | 0.32 |
| History of smoking | 58 (12.8%) | 657 (10.6%) | 715 (10.8%) | 0.22 |
| Part D Enrollment | 0.0012 | |||
| Enrolled: total copayment subsidy | 54 (11.8%) | 478 (7.6%) | 532 (7.9%) | |
| Enrolled: moderate copayment subsidy | 167 (36.4%) | 2017 (32.2%) | 2184 (32.5%) | |
| Enrolled: small or no copayment subsidy | 152 (33.1%) | 2366 (37.7%) | 2518 (37.4%) | |
| Not enrolled | 86 (18.7%) | 1407 (22.4%) | 1493 (22.2%) | |
| Health history | ||||
| Hypertension | 427 (94.1%) | 5761 (93.4%) | 6188 (93.4%) | 0.62 |
| Diabetes mellitus | 327 (71.2%) | 4783 (76.3%) | 5110 (76.0%) | 0.012 |
| Cancer | 61 (13.3%) | 955 (15.2%) | 1016 (15.1%) | 0.32 |
| Comorbidity Index | 7 (3, 12) | 8 (5, 11) | 8 (5, 11) | 0.021 |
| Cardiovascular hospitalizations past 6m | 41 (8.9%) | 675 (10.8%) | 716 (10.6%) | 0.22 |
| Functional Limitation4 | 63 (13.9%) | 816 (13.2%) | 879 (13.3%) | 0.72 |
| Recent clinical characteristics | ||||
| BMI (kg/m^2) | 29.6 (25.4, 35.0) | 28.6 (24.6, 34.0) | 28.6 (24.6, 34.1) | 0.0041 |
| PTH (pg/ml) | 545 (419, 684) | 443 (355, 546) | 447 (357, 556) | <0.0011 |
| Phosphorus (mg/dl) | 5.4 (4.6, 6.5) | 5.3 (4.6, 6.2) | 5.3 (4.6, 6.2) | 0.11 |
| Albumin Corrected Calcium (mg/dl) | 9.2 (8.8, 9.6) | 9.2 (8.9, 9.6) | 9.2 (8.9, 9.6) | 0.21 |
| Dialysis characteristics | ||||
| Dialysis vintage (years) | 29.6 (25.4, 35.0) | 28.6 (24.6, 34.0) | 28.6 (24.6, 34.1) | 0.0041 |
| Primary cause of kidney failure5 | 0.22 | |||
| Diabetes mellitus | 225 (49.0%) | 3325 (53.0%) | 3550 (52.8%) | |
| Hypertension | 135 (29.4%) | 1772 (28.3%) | 1907 (28.3%) | |
| Glomerulonephritis | 42 (9.2%) | 417 (6.7%) | 459 (6.8%) | |
| Other cause | 57 (12.4%) | 747 (11.9%) | 804 (12.0%) | |
| Mean dialysis time (minutes) | 213.4 (195.5, 237.9) | 210.7 (193.5, 233.9) | 210.9 (193.6, 234.3) | 0.011 |
| Dialysate Calcium (mEq/l)6 | 0.52 | |||
| <2.5(%) | 137 (29.8%) | 1972 (31.5%) | 2109 (31.4%) | |
| ≥2.5(%) | 322 (70.2%) | 4295 (68.5%) | 4617 (68.6%) | |
| Kt/V | 1.5 (1.4, 1.6) | 1.5 (1.4, 1.7) | 1.5 (1.4, 1.7) | <0.0011 |
| Primary vascular access | 0.12 | |||
| Fistula | 301 (65.6%) | 3822 (61.0%) | 4123 (61.3%) | |
| Graft | 93 (20.3%) | 1486 (23.7%) | 1579 (23.5%) | |
| Catheter | 65 (14.2%) | 960 (15.3%) | 1025 (15.2%) | |
| Secular trends | ||||
| Year of incident kidney failure | <0.0012 | |||
| 2006 | 72 (15.7%) | 944 (15.1%) | 1016 (15.1%) | |
| 2007 | 60 (13.1%) | 1126 (18.0%) | 1186 (17.6%) | |
| 2008 | 81 (17.6%) | 1323 (21.1%) | 1404 (20.9%) | |
| 2009 | 77 (16.8%) | 1064 (17.0%) | 1141 (17.0%) | |
| 2010 | 62 (13.5%) | 701 (11.2%) | 763 (11.3%) | |
| 2011 | 46 (10.0%) | 465 (7.4%) | 511 (7.6%) | |
| 2012 | 33 (7.2%) | 443 (7.1%) | 476 (7.1%) | |
| 2013 | 23 (5.0%) | 186 (3.0%) | 209 (3.1%) | |
| Year of trial eligibility | <0.0012 | |||
| 2009 | 71 (15.5%) | 1320 (21.1%) | 1391 (20.7%) | |
| 2010 | 68 (14.8%) | 1474 (23.5%) | 1542 (22.9%) | |
| 2011 | 89 (19.4%) | 840 (13.4%) | 929 (13.8%) | |
| 2012 | 75 (16.3%) | 886 (14.1%) | 961 (14.3%) | |
| 2013 | 55 (12.0%) | 818 (13.1%) | 873 (13.0%) | |
| 2014 | 101 (22.0%) | 930 (14.8%) | 1031 (15.3%) | |
| Time from first trial eligibility (years) | 0.7 (0.1, 1.9) | 0.7 (0.0, 1.8) | 0.7 (0.0, 1.8) | 0.21 |
Abbreviations: PTH=parathyroid hormone; BMI=body mass index
p-value calculated for two-sided Kruskal Wallis test with 5% significance
p-value calculated for two-sided Chi-square test with 5% significance
When examining unadjusted mean PTH levels in the cohort during the two-year follow-up, both cohorts had lower mean PTH levels in the first 3–4 months post-eligibility, with unadjusted mean and median PTH levels higher in both cohorts thereafter (Figure 5a). An exploration of the titrations after baseline demonstrated that the vitamin D sterol titration was more common than cinacalcet titration in both arms, although cinacalcet titration was much more common in the cinacalcet-favoring arm than the vitamin D-favoring arm (Figure 5b).
Figure 5. Levels of Parathyroid Hormone and Medication Titrations Over Time in Patients Participating in the Agent Trial.
In Panel A, the unadjusted achieved parathyroid hormone (PTH) level in the cohort over 24 months on study is demonstrated. Patients following a cinacalcet-favoring approach are depicted in solid lines with vitamin D-favoring approach depicted in dashed lines. Bars represent 95% confidence intervals around each monthly mean level (depicted in black). For comparison, median in each group is depicted in red. Not all patients are represented in each summary distribution due to censoring over time with the number included presenting in the table below the X-axis. Missing PTH values in any given month are carried forward from the most recent measure. In Panel B, loess smoothed functions of proportions of patients with medication titrations post-baseline are depicted in each group over follow-up with their 95% confidence intervals. Solid lines depict any titration of vitamin D sterols. Dashed lines depict any titration of cinacalcet. Blue lines in each set reflect the cinacalcet-favoring treatment group and red lines reflect the vitamin D-favoring treatment group. Time 0 is 30 days after the eligible PTH, when follow-up begins. The exposure assignment was based on medication titration occurring in the 30 day exposure assessment window that precedes this follow-up period.
Amongst all patients included in the Agent Trial, there were 745 deaths occurring in 27.3% of patients and 25.5% of trials. Overall 1,128 patients (41.4%) were hospitalized for a cardiovascular cause and these occurred in 1264, or 43.3% of trials. The composite outcome was experienced by 1,474 (54.1%) patients and occurred in 1667, or 57.1% of trials. For the intention-to-treat analysis, the estimated hazard ratio for death was 0.79 (95% CI: 0.62 to 0.99) for the cinacalcet-favoring versus vitamin D-favoring arm (Table 4). However, hazard ratios for secondary endpoints were null. Standardized event rates did not show consistent direction of effect (Figure S3).
Table 4.
Pooled Logistic Regression Estimation of Hazard Ratio for Death, First Cardiac Hospitalization, and a Composite Over 24 Months for Cinacalcet-Favoring Compared to Vitamin D-Favoring Observations in the Agent Trial
| Intention-to-Treat | Per-protocol | ||
|---|---|---|---|
| Unadjusted | Adjusted | Adjusted | |
| Outcome | Hazard Ratio (95% CI) | Hazard Ratio (95% CI) | Hazard Ratio (95% CI) |
| Death | 0.82 (0.65, 1.00) | 0.79 (0.62, 0.99) | 1.01 (0.66, 2.77) |
| Cardiac Hospitalization | 0.95 (0.81, 1.11) | 1.01 (0.84, 1.20) | 1.07 (0.73, 1.79) |
| Composite Death and Cardiac Hospitalizations | 0.89 (0.77, 1.03) | 0.93 (0.79, 1.09) | 1.01 (0.72, 1.69) |
Note: Hazard ratios reflect the cinacalcet-favoring compared to vitamin D-favoring group
Prescription patterns over time indicated a steadily increasing rate of going off protocol, particularly in the first 6 months of follow-up. By 6 months about 20% of patients in the cinacalcet-favoring arm had gone off protocol, while about 20% of patients in the vitamin D-favoring arm went off protocol by 24 months (Figure S4). For the per-protocol analysis, there were 630 deaths occurring in 23.1% of patients and 17.5% of trials, 974 hospitalizations with cardiac cause occurring 35.7% of patients and 29.9% of trials, and 1273 composite events occurring in 46.7% of patients and 51.2% of trials. The estimated per protocol hazard ratio for death for the cinacalcet-favoring arm vs. the vitamin D-favoring arm was 1.01 (95% CI: 0.66 to 2.77; Table 4). Additional per-protocol effects are presented in Table 4. There were no meaningful differences in the treatment effects by age (Table S5) or over different follow-up time (Table S6).
Outcome analyses for the Agent Trial were similar when the sample was limited to patients with incident in-center hemodialysis starting in 2009 (Table S3), when only 90-days of look-back for Medicare Primary Payer was required (Table S4), and when we required that a single titration could only be associated with one trial (Table S7).
Discussion
In these two trial emulation studies evaluating the comparative effectiveness of different treatment approaches to management of secondary hyperparathyroidism in patients treated with maintenance hemodialysis, we report 2 major findings. First, in an emulation of a PTH Target Trial, we found that patients treated with a lower target approach had lower rates of overall mortality and cardiovascular hospitalizations compared to patients treated with a higher target approach. The lower target approach was defined by up-titration of vitamin D sterols and/or cinacalcet within 30 days of the first PTH between 300–600 pg/ml, and the higher target approach was defined by delayed up-titration of these agents. Providers have been targeting higher ranges of PTH since CKD-MBD guidelines were relaxed in 2009 as evidenced by rising PTH levels among patients treated with hemodialysis in the US.20 Our findings suggest that this trend, allowing PTH to rise higher before titrating PTH controlling medication, may worsen outcomes in patients with kidney failure. In the second trial, we did not identify any clear benefit or harm of an approach characterized by early addition of cinacalcet, dubbed cinacalcet-favoring, or an approach in which vitamin D sterols are favored although power was low. These results are generated from an observational design and thus residual confounding by treatment indication is possible despite careful design and covariate adjustment. Based on our findings, the hypothesis that more proactive use of PTH controlling medications to lower PTH levels and reduce mortality and cardiovascular disease should be rigorously tested in a follow-up pragmatic PTH target randomized controlled trial to confirm these findings.
The Target Trial emulation defined lower target based on titration of cinacalcet or vitamin D sterols at the PTH range of 300–600 pg/ml. Patients treated with this approach exhibited generally lower PTH over the first year of follow up compared to patients treated with delayed titration, or a higher target approach. In essence, these more timely initial medication titrations revealed a provider preference, or intention, to keep PTH lower within the guideline suggested target range. Subsequent titration patterns revealed that patients in the lower PTH arm had more medication titration for PTH generally, including upward and downward titration after follow-up. These subsequent downward titrations may not reflect non-adherence to the protocol because downward titration in response to over-suppressed PTH or side effects would be allowed.
Most prior studies in the field have focused on the association between achieved PTH levels and patient outcomes. Their results have shown a range of findings, including no association, graded risk with higher PTH, or a U-shaped association with increased risk at both higher and lower PTH level.21, 22 However, these results do not clearly reflect the estimated effect of randomized intention-to-treat in a trial because the achieved PTH level is influenced by many factors other than the randomized target, including the severity of the patient’s disease and their refractoriness or adherence to therapy.
The generally null results of our Agent Trial emulation should be interpreted in the context of prior studies in the field. In 2012, the EVOLVE trial demonstrated a non-significant 7% reduction in the hazard of a cardiovascular and mortality composite outcome for patients treated with cinacalcet versus placebo on top of standard of care.23 These effects were marginally significant when adjusted for baseline imbalance in patient characteristics in post-hoc analyses and thus have been viewed as inconclusive by many in the field. Our results support a lack of treatment effect, but are not directly comparable. In EVOLVE patients randomized to cinacalcet achieved lower PTH and lower serum calcium over follow-up compared to those randomized to placebo. In our study we observed higher PTH at baseline and over follow-up among the cohort in the cinacalcet-favoring arm. Although we do not interpret these patients as fully refractory to vitamin D sterols because our inclusion requirements required PTH still within guideline targets and no contraindications for vitamin D, such as hypercalcemia, this may suggest more refractory disease among patients assigned to the cinacalcet arm in our study. We adjusted for baseline difference in PTH across groups but it is possible that these adjustments could not fully remove these effects and may not recapitulate the lower PTH and calcium achieved with cinacalcet treatment over follow-up. We also had fewer patients following the cinacalcet-favoring approach in this study leading to low power, perhaps due to the use of incident dialysis patients.7 With these limitations in mind and the wide confidence intervals for the treatment effects, we consider these analyses inconclusive, unable to demonstrate a clear benefit or equivalence of the two approaches. It is important to note that we did find a modestly lower hazard of mortality over 24 months with the cinacalcet-favoring approach, but we interpret this cautiously because per-protocol effects and standardized event rates revealed different results, and we did not see any associations with cardiovascular hospitalization or the composite outcome.
Our approach uses real-world clinical practice data and evaluates important clinical outcomes in a rigorously designed causal framework to emulate results of a comparable clinical trial. Simulation studies demonstrate that these types of careful emulation studies can overcome many biases in observational studies that may exhibit flaws in counterfactual logic.15, 24 Using a large and detailed dataset we were able to precisely isolate a specific clinical context in which a provider faces a randomizable treatment choice. Ultimately these restrictions designed to mimic the specific inclusion and exclusion criteria of a trial caused substantial loss of sample size and power. For this reason, even in a large dataset we may have been unable to detect important and precise effect estimates or prove equivalence. Our data included many important clinical markers such as evolving laboratory values and precise information about dialysis treatments to control confounding. However, we were not able to observe clinical characteristics prior to dialysis initiation or outside of the dialysis environment that could lead to confounding. Also, analyses include competing events that may lead to censoring. For instance, patients receiving kidney transplant are censored from analyses. In analyses of cardiovascular hospitalization, patients who die are censored. Our methods assume that patients who remain in the analysis provide a reliable estimate of the hazard for all patients in the study, including those censored.
Several factors affect the potential generalizability of our results. We required a look-back period to allow patients to acquire Medicare coverage and to accumulate Medicare claims to assess comorbidity. Many potential patients in the Target Trial had a first PTH in the range of 300–600 pg/ml shortly after dialysis initiation, and thus were excluded. Furthermore, our requirement that patients have Medicare as a primary payer also limits generalizability to this group for both Trials.
In conclusion, differences in PTH management focused on greater treatment intensity to lower versus higher PTH targets may lower mortality and cardiovascular disease in hemodialysis and is an ideal contrast for future clinical trials. With an approximately 25–30% reduction in cardiovascular disease and death in intention to treat analyses, these potential effects are clinically meaningful. Marginally smaller estimated per-protocol effects warrant further investigation and concerted efforts to maintain protocol adherence in a subsequent clinical trial. These results are derived from an observational trial emulation that still may be subject to confounding, but should help prioritize and design a confirmatory, pragmatic randomized controlled PTH target trial.
Supplementary Material
Supplementary Material
Supplementary File (PDF)
Figure S1. Standardized Event Curves- Target Trial
Figure S2. Adherence- Target Trial
Figure S3. Standardized Event Curves- Agent Trial
Figure S4. Adherence- Agent Trial
Item S1: Detailed Methods
Item S2: Detailed Rationale for Emulation Design
Item S3: Drug Ascertainment and Dose Conversion
Item S4: Detailed Computable Definitions of Key Variables
Table S1. Age-Stratification- Target Trial
Table S2. Hazard Ratios at 6, 9, 12 and 18 m- Target Trial
Table S3. Sensitivity in Incident Cohort- Both Trials
Table S4. Sensitivity to Look-Back Period- Both Trials
Table S5. Age-Stratification- Agent Trial
Table S6. Hazard Ratios at 6, 9, 12 and 18 m- Agent Trial
Table S7. Sensitivity to Removing Closely Linked Trials
Acknowledgments:
The authors are grateful to the staff and patients of Dialysis Clinic, Inc.
Support:
This study was supported by R01DK111952 from the National Institute of Diabetes and Digestive and Kidney Diseases. Additional support was provided in part by National Institute on Aging Award Number K76AG059930 (RH), National Center for Advancing Translational Sciences of the National Institutes of Health Award Number UL1TR002553, and the ASN Foundation for Kidney Research (RH). Neither the sponsors, nor Dialysis Clinic, Inc had a deciding role in the study design, analysis, interpretation of the data, writing of the report, or the decision to submit the report for publication.
Footnotes
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Financial Disclosure:The authors declare that they have no relevant financial interests.
Other Disclosures: Dr Scialla serves as Deputy Editor for the American Journal of Kidney Diseases.
Disclaimer: The manuscript reflects the interpretation and opinions of the authors and is not expressly endorsed by the National Institutes of Health, the National Institute of Diabetes and Digestive and Kidney Diseases, National Institute on Aging, or DCI. The data reported here have been supplied by the United States Renal Data System (USRDS). The interpretation and reporting of these data are the responsibility of the author(s) and in no way should be seen as an official policy or interpretation of the U.S. government.
Prior Presentation: Presented in part at the American Society of Nephrology 2022 Kidney Week; November 3, 2022; Orlando, FL.
References
- 1.Scialla JJ. Evidence basis for integrated management of mineral metabolism in patients with end-stage renal disease. Curr Opin Nephrol Hypertens. Jul 2018;27(4):258–267. doi: 10.1097/MNH.0000000000000417 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Ritz E, Krempien B, Mehls O, Malluche H. Skeletal abnormalities in chronic renal insufficiency before and during maintenance hemodialysis. Kidney Int. 1973/08/01/ 1973;4(2):116–127. doi: 10.1038/ki.1973.90 [DOI] [PubMed] [Google Scholar]
- 3.Lunyera J, Scialla JJ. Update on Chronic Kidney Disease Mineral and Bone Disorder in Cardiovascular Disease. Semin Nephrol. Nov 2018;38(6):542–558. doi: 10.1016/j.semnephrol.2018.08.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Scialla JJ, Wolf M. Roles of phosphate and fibroblast growth factor 23 in cardiovascular disease. Nat Rev Nephrol. May 2014;10(5):268–78. doi: 10.1038/nrneph.2014.49 [DOI] [PubMed] [Google Scholar]
- 5.Kidney Disease: Improving Global Outcomes CKD-MBD Update Work Ggroup. KDIGO 2017 Clinical Practice Guideline Update for the Diagnosis, Evaluation, Prevention, and Treatment of Chronic Kidney Disease-Mineral and Bone Disorder (CKD-MBD). Kidney Int Suppl (2011). Jul 2017;7(1):1–59. doi: 10.1016/j.kisu.2017.04.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Eknoyan G, Levin A, Levin NW. Bone metabolism and disease in chronic kidney disease. Am J Kidney Dis. Oct 2003;42(Suppl 3):S1–201. doi: 10.1016/S0272-6386(03)00905-3 [DOI] [PubMed] [Google Scholar]
- 7.Hall R, Platt A, Wilson J, et al. Trends in Mineral Metabolism Treatment Strategies in Patients Receiving Hemodialysis in the United States. Clin J Am Soc Nephrol. 2020;15(11):1603–1613. doi: 10.2215/cjn.04350420 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Fu EL, Evans M, Clase CM, et al. Stopping Renin-Angiotensin System Inhibitors in Patients with Advanced CKD and Risk of Adverse Outcomes: A Nationwide Study. J Am Soc Nephrol. Feb 2021;32(2):424–435. doi: 10.1681/ASN.2020050682 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Xie Y, Bowe B, Gibson AK, et al. Comparative Effectiveness of SGLT2 Inhibitors, GLP-1 Receptor Agonists, DPP-4 Inhibitors, and Sulfonylureas on Risk of Kidney Outcomes: Emulation of a Target Trial Using Health Care Databases. Diabetes Care. Nov 2020;43(11):2859–2869. doi: 10.2337/dc20-1890 [DOI] [PubMed] [Google Scholar]
- 10.Hernán MA, Robins JM. Using Big Data to Emulate a Target Trial When a Randomized Trial Is Not Available. Am J Epidemiol. Apr 15 2016;183(8):758–64. doi: 10.1093/aje/kwv254 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.KDIGO clinical practice guideline for the diagnosis, evaluation, prevention, and treatment of Chronic Kidney Disease-Mineral and Bone Disorder (CKD-MBD). Kidney Int Suppl. Aug 2009;(113):S1–130. doi: 10.1038/ki.2009.188 [DOI] [PubMed] [Google Scholar]
- 12.Saran R, Robinson B, Abbott KC, et al. US Renal Data System 2019 Annual Data Report: epidemiology of kidney disease in the United States. Am J Kidney Dis. 2020;75(1)(suppl 1):S1–S64. doi: 10.1053/j.ajkd.2019.09.003 [DOI] [PubMed] [Google Scholar]
- 13.D’Agostino RB, Lee ML, Belanger AJ, Cupples LA, Anderson K, Kannel WB. Relation of pooled logistic regression to time dependent Cox regression analysis: the Framingham Heart Study. Stat Med. Dec 1990;9(12):1501–15. doi: 10.1002/sim.4780091214 [DOI] [PubMed] [Google Scholar]
- 14.Liu J, Huang Z, Gilbertson DT, Foley RN, Collins AJ. An improved comorbidity index for outcome analyses among dialysis patients. Kidney Int. Jan 2010;77(2):141–51. doi: 10.1038/ki.2009.413 [DOI] [PubMed] [Google Scholar]
- 15.Danaei G, Rodríguez LA, Cantero OF, Logan R, Hernán MA. Observational data for comparative effectiveness research: an emulation of randomised trials of statins and primary prevention of coronary heart disease. Stat Methods Med Res. Feb 2013;22(1):70–96. doi: 10.1177/0962280211403603 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Robins JM, Hernán MÁ, Brumback B. Marginal Structural Models and Causal Inference in Epidemiology. Epidemiol. 2000;11(5):550–560. [DOI] [PubMed] [Google Scholar]
- 17.Petito LC, García-Albéniz X, Logan RW, et al. Estimates of Overall Survival in Patients With Cancer Receiving Different Treatment Regimens: Emulating Hypothetical Target Trials in the Surveillance, Epidemiology, and End Results (SEER)–Medicare Linked Database. JAMA Netw Open. 2020;3(3):e200452–e200452. doi: 10.1001/jamanetworkopen.2020.0452 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Hernán MA. The hazards of hazard ratios. Epidemiol. Jan 2010;21(1):13–5. doi: 10.1097/EDE.0b013e3181c1ea43 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Parfrey PS, Drüeke TB, Block GA, et al. The Effects of Cinacalcet in Older and Younger Patients on Hemodialysis: The Evaluation of Cinacalcet HCl Therapy to Lower Cardiovascular Events (EVOLVE) Trial. Clin J Am Soc Nephrol. May 7 2015;10(5):791–9. doi: 10.2215/cjn.07730814 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Tentori F, Zepel L, Fuller DS, et al. The DOPPS Practice Monitor for US Dialysis Care: PTH Levels and Management of Mineral and Bone Disorder in US Hemodialysis Patients. Am J Kidney Dis. 2015;66(3):536–539. doi: 10.1053/j.ajkd.2015.07.011 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Tentori F, Blayney MJ, Albert JM, et al. Mortality risk for dialysis patients with different levels of serum calcium, phosphorus, and PTH: the Dialysis Outcomes and Practice Patterns Study (DOPPS). Am J Kidney Dis. Sep 2008;52(3):519–30. doi: 10.1053/j.ajkd.2008.03.020 [DOI] [PubMed] [Google Scholar]
- 22.Palmer SC, Hayen A, Macaskill P, et al. Serum levels of phosphorus, parathyroid hormone, and calcium and risks of death and cardiovascular disease in individuals with chronic kidney disease: a systematic review and meta-analysis. JAMA. Mar 16 2011;305(11):1119–27. doi: 10.1001/jama.2011.308 [DOI] [PubMed] [Google Scholar]
- 23.EVOLVE Trial Investigators, Chertow GM, Block GA, et al. Effect of cinacalcet on cardiovascular disease in patients undergoing dialysis. N Engl J Med. Dec 27 2012;367(26):2482–94. doi: 10.1056/NEJMoa1205624 [DOI] [PubMed] [Google Scholar]
- 24.Dickerman BA, García-Albéniz X, Logan RW, Denaxas S, Hernán MA. Avoidable flaws in observational analyses: an application to statins and cancer. Nat Med. 2019/10/01 2019;25(10):1601–1606. doi: 10.1038/s41591-019-0597-x [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplementary Material
Supplementary File (PDF)
Figure S1. Standardized Event Curves- Target Trial
Figure S2. Adherence- Target Trial
Figure S3. Standardized Event Curves- Agent Trial
Figure S4. Adherence- Agent Trial
Item S1: Detailed Methods
Item S2: Detailed Rationale for Emulation Design
Item S3: Drug Ascertainment and Dose Conversion
Item S4: Detailed Computable Definitions of Key Variables
Table S1. Age-Stratification- Target Trial
Table S2. Hazard Ratios at 6, 9, 12 and 18 m- Target Trial
Table S3. Sensitivity in Incident Cohort- Both Trials
Table S4. Sensitivity to Look-Back Period- Both Trials
Table S5. Age-Stratification- Agent Trial
Table S6. Hazard Ratios at 6, 9, 12 and 18 m- Agent Trial
Table S7. Sensitivity to Removing Closely Linked Trials