Parathyroid Hormone and Cardiovascular Disease Events
Parathyroid Hormone and Cardiovascular Disease Events
We conducted this systematic review and meta-analysis according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement and the Meta-analysis of Observational Studies in epidemiology (MOOSE) guideline. A specialized librarian from the VU University Medical Center Amsterdam conducted a systematic and comprehensive database search of MEDLINE (1948 to January 2012) and Embase (1947 to January 2012). In the search strategy we combined text words, Medical Subject Headings, and Emtree terms without language restriction. Search criteria included 'parathyroid hormone' and 'cardiovascular events' and their corresponding index words using the special features ("explosion" and "truncation") in titles and/or abstracts. Only human studies were included. A complete overview of the search criteria is attached (see online Appendix Supplement 1). Unpublished material was not considered. The final cut-off date for literature inclusion was January 24, 2012. In addition, we verified the search strategy by hand-searching the reference lists of primary studies, review articles, and clinical guidelines. Email-alerts with newly published articles from MEDLINE were checked until October 1, 2012.
Two authors (AJvB and IR) independently screened articles based on title and abstract and selected articles according to inclusion and exclusion criteria, without blinding of the study authors. Inclusion criteria were the following: (1) serum or plasma PTH was the determinant; (2) cardiovascular events (fatal and/or nonfatal), intermediate outcomes, cardiac morphological, or cardiac biomarkers were the outcome measure; (3) the authors report data of an original prospective quantitative cohort study (i.e. no review, commentary, case reports, editorial, or qualitative research); (4) study performed in adults ≥18 year; (5) published in English. We excluded studies performed in patient populations, that is, all patients were on dialysis, all patients had an estimated glomerular filtration rate below 60 mL/min, or if patients underwent parathyroidectomy. Differences between authors were resolved by consensus or by consultation of an additional reviewer (IAB).
In case of insufficient information based on title or abstract, the study was taken to the next step in the review process. In this stage, 2 authors independently inspected full-text articles. In the final step, we re-inspected the full text articles and excluded articles based on non-prospective study design. Of the remaining prospective studies we recorded first author; years of follow-up; country of origin; name of the study; number of participants/cases; participants' characteristics; determination of outcome; PTH concentrations; and outcome measures. Information on kidney function and parathyroid disorders, including primary hyperparathyroidism was also recorded.
We developed a quality checklist to determine the quality of the eligible studies based on the PRISMA Statement and the MOOSE guideline in combination with a previously published quality checklist for observational studies (see online Appendix Supplement 2). Studies were scored based on 9 quality criteria on a binary scale; including description of characteristics of the study population, assessment of exposure and outcome, confounding and potential flaws. The quality of each study was independently assessed by AJvB and IR and scores were compared for each item of the checklist. The scores were summed and quality was considered poor (0–4), moderate (5–6), or good (7–9). Studies with potential flaws or rated as poor quality were not included.
Results reported as count data were presented for total CVD events, fatal CVD events and non-fatal CVD events: adjusted HR, OR or regression coefficients (β) and 95% CIs. We extracted the results of the highest versus lowest PTH concentrations and used the lowest PTH category as the reference category. If the study reported more than 1 estimate, only the result of the largest PTH difference was included. For example, the study by Hagström reported the results of quartile 4 versus quartile 1 (HR: 1.83 [1.10–3.04]) and the results of a threshold model quartile 4 versus quartile 1 to 3 (HR: 1.98 [1.35–2.92]). We only took into account the result of quartile 4 versus quartile 1. We transformed risk estimates by taking their natural logarithms and calculated the standard errors as follows: (Ln upper limit – Ln HR)/1.96. We weighted the natural logarithm of the risk estimates by generic inverse variance to account for the sample size and distribution of the included studies. For 3 studies, descriptive data could not be extracted due to lack of information or conflicting information. In one study, we confirmed data by the original authors and in case of 2 studies we could not obtain additional information and presented it as unknown.
The results of the included studies were pooled and meta-analyses were carried out using fixed and random-effects models. Different effects models were used to investigate whether heterogeneity influenced the pooled estimate. This resulted in multiple forest plots, a funnel plot and results are presented as pooled HRs with 95% CI. Potential publication bias was evaluated by visual inspection of a funnel plot. We presented the results of random-effects model analyses, because random-effects models assume that the associations of different prospective studies differ, but follow a normal distribution. With increasing heterogeneity the estimates between fixed and random models will be more divergent. We evaluated heterogeneity between-study variance by τ, χ and I. τ is the estimated standard deviation of underlying effects across studies. The χ symbolizes the difference between the observed and expected result. A low P value provides evidence of heterogeneity between studies (variation in estimates beyond chance). I values represent the proportion of total variation attributable to heterogeneity rather than chance whereby 0% is no observed heterogeneity and 100% maximal heterogeneity.
A priori sensitivity analyses were defined to evaluate the stability of the pooled estimates and to examine changes in results after excluding specific studies. The subgroup analyses were preplanned for: length of follow-up >6 year, subjects with no cardiac diseases at baseline and limited to studies with PTH categories as quartiles. Meta-analyses were performed by Review Manager 5.1 (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark; http://www.cc-ims.net/RevMan). All P values were 2-tailed (α = .05).
This work was supported by the Royal Netherlands Academy of Arts and Sciences. The authors are solely responsible for the design and conduct of this study, all analyses, the drafting and editing of the manuscript, and its final content.
Methods
Search Strategy
We conducted this systematic review and meta-analysis according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement and the Meta-analysis of Observational Studies in epidemiology (MOOSE) guideline. A specialized librarian from the VU University Medical Center Amsterdam conducted a systematic and comprehensive database search of MEDLINE (1948 to January 2012) and Embase (1947 to January 2012). In the search strategy we combined text words, Medical Subject Headings, and Emtree terms without language restriction. Search criteria included 'parathyroid hormone' and 'cardiovascular events' and their corresponding index words using the special features ("explosion" and "truncation") in titles and/or abstracts. Only human studies were included. A complete overview of the search criteria is attached (see online Appendix Supplement 1). Unpublished material was not considered. The final cut-off date for literature inclusion was January 24, 2012. In addition, we verified the search strategy by hand-searching the reference lists of primary studies, review articles, and clinical guidelines. Email-alerts with newly published articles from MEDLINE were checked until October 1, 2012.
Study Selection
Two authors (AJvB and IR) independently screened articles based on title and abstract and selected articles according to inclusion and exclusion criteria, without blinding of the study authors. Inclusion criteria were the following: (1) serum or plasma PTH was the determinant; (2) cardiovascular events (fatal and/or nonfatal), intermediate outcomes, cardiac morphological, or cardiac biomarkers were the outcome measure; (3) the authors report data of an original prospective quantitative cohort study (i.e. no review, commentary, case reports, editorial, or qualitative research); (4) study performed in adults ≥18 year; (5) published in English. We excluded studies performed in patient populations, that is, all patients were on dialysis, all patients had an estimated glomerular filtration rate below 60 mL/min, or if patients underwent parathyroidectomy. Differences between authors were resolved by consensus or by consultation of an additional reviewer (IAB).
In case of insufficient information based on title or abstract, the study was taken to the next step in the review process. In this stage, 2 authors independently inspected full-text articles. In the final step, we re-inspected the full text articles and excluded articles based on non-prospective study design. Of the remaining prospective studies we recorded first author; years of follow-up; country of origin; name of the study; number of participants/cases; participants' characteristics; determination of outcome; PTH concentrations; and outcome measures. Information on kidney function and parathyroid disorders, including primary hyperparathyroidism was also recorded.
Quality Assessment
We developed a quality checklist to determine the quality of the eligible studies based on the PRISMA Statement and the MOOSE guideline in combination with a previously published quality checklist for observational studies (see online Appendix Supplement 2). Studies were scored based on 9 quality criteria on a binary scale; including description of characteristics of the study population, assessment of exposure and outcome, confounding and potential flaws. The quality of each study was independently assessed by AJvB and IR and scores were compared for each item of the checklist. The scores were summed and quality was considered poor (0–4), moderate (5–6), or good (7–9). Studies with potential flaws or rated as poor quality were not included.
Statistical Methods
Results reported as count data were presented for total CVD events, fatal CVD events and non-fatal CVD events: adjusted HR, OR or regression coefficients (β) and 95% CIs. We extracted the results of the highest versus lowest PTH concentrations and used the lowest PTH category as the reference category. If the study reported more than 1 estimate, only the result of the largest PTH difference was included. For example, the study by Hagström reported the results of quartile 4 versus quartile 1 (HR: 1.83 [1.10–3.04]) and the results of a threshold model quartile 4 versus quartile 1 to 3 (HR: 1.98 [1.35–2.92]). We only took into account the result of quartile 4 versus quartile 1. We transformed risk estimates by taking their natural logarithms and calculated the standard errors as follows: (Ln upper limit – Ln HR)/1.96. We weighted the natural logarithm of the risk estimates by generic inverse variance to account for the sample size and distribution of the included studies. For 3 studies, descriptive data could not be extracted due to lack of information or conflicting information. In one study, we confirmed data by the original authors and in case of 2 studies we could not obtain additional information and presented it as unknown.
The results of the included studies were pooled and meta-analyses were carried out using fixed and random-effects models. Different effects models were used to investigate whether heterogeneity influenced the pooled estimate. This resulted in multiple forest plots, a funnel plot and results are presented as pooled HRs with 95% CI. Potential publication bias was evaluated by visual inspection of a funnel plot. We presented the results of random-effects model analyses, because random-effects models assume that the associations of different prospective studies differ, but follow a normal distribution. With increasing heterogeneity the estimates between fixed and random models will be more divergent. We evaluated heterogeneity between-study variance by τ, χ and I. τ is the estimated standard deviation of underlying effects across studies. The χ symbolizes the difference between the observed and expected result. A low P value provides evidence of heterogeneity between studies (variation in estimates beyond chance). I values represent the proportion of total variation attributable to heterogeneity rather than chance whereby 0% is no observed heterogeneity and 100% maximal heterogeneity.
A priori sensitivity analyses were defined to evaluate the stability of the pooled estimates and to examine changes in results after excluding specific studies. The subgroup analyses were preplanned for: length of follow-up >6 year, subjects with no cardiac diseases at baseline and limited to studies with PTH categories as quartiles. Meta-analyses were performed by Review Manager 5.1 (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark; http://www.cc-ims.net/RevMan). All P values were 2-tailed (α = .05).
This work was supported by the Royal Netherlands Academy of Arts and Sciences. The authors are solely responsible for the design and conduct of this study, all analyses, the drafting and editing of the manuscript, and its final content.
