SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. AUTHOR CONTRIBUTIONS
  8. REFERENCES

Objective

To determine the magnitude of risk of cardiovascular mortality in patients with rheumatoid arthritis (RA) compared with the general population through a meta-analysis of observational studies.

Methods

We searched Medline, EMBase, and Lilacs databases from their inception to July 2005. Observational studies that met the following criteria were assessed by 2 researchers: 1) prespecified RA definition, 2) clearly defined cardiovascular disease (CVD) outcome, including ischemic heart disease (IHD) and cerebrovascular accidents (CVAs), and 3) reported standardized mortality ratios (SMRs) and 95% confidence intervals (95% CIs). We calculated weighted–pooled summary estimates of SMRs (meta-SMRs) for CVD, IHD, and CVAs using the random-effects model, and tested for heterogeneity using the I2 statistic.

Results

Twenty-four studies met the inclusion criteria, comprising 111,758 patients with 22,927 cardiovascular events. Overall, there was a 50% increased risk of CVD death in patients with RA (meta-SMR 1.50, 95% CI 1.39–1.61). Mortality risks for IHD and CVA were increased by 59% and 52%, respectively (meta-SMR 1.59, 95% CI 1.46–1.73 and meta-SMR 1.52, 95% CI 1.40–1.67, respectively). We identified asymmetry in the funnel plot (Egger's test P = 0.002), as well as significant heterogeneity in all main analyses (P < 0.0001). Subgroup analyses showed that inception cohort studies (n = 4, comprising 2,175 RA cases) were the only group that did not show a significantly increased risk for CVD (meta-SMR 1.19, 95% CI 0.86–1.68).

Conclusion

Published data indicate that CVD mortality is increased by ∼50% in RA patients compared with the general population. However, we found that study characteristics may influence the estimate.


INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. AUTHOR CONTRIBUTIONS
  8. REFERENCES

Rheumatoid arthritis (RA) is a chronic inflammatory disease that leads to progressive joint deformity, disability, and arguably to premature death (1–26). Most studies of mortality in patients with RA have found increased mortality rates compared with the general population (1–15), and the majority suggest that one-third to one-half of the premature deaths in RA patients are due to increased cardiovascular disease (CVD), including ischemic heart disease (IHD) and cerebrovascular accidents (CVAs). However, there has been wide variation in reported standardized mortality ratios (SMRs), including studies detecting neither increased overall mortality (20–23) nor CVD mortality (19–26).

Some recent studies reported improvement in survival and suggested that these improvements may be related to earlier diagnosis and the use of more aggressive and newer antirheumatic treatment regimens (17, 27–30). As a result, CVD has become a relevant long-term end point for RA, especially in studies of therapy.

Previous studies evaluating RA mortality have shown that those attending hospitals for treatment have a reduced life expectancy compared with control populations (1, 2, 4, 21, 22, 30–33). This suggests that study design might also explain some of the differences in mortality rates reported in patients with RA.

Increasingly, it has been recognized that inflammation plays an important role in atherosclerosis (34–36), making the evaluation of death from cardiovascular causes and RA all the more important, because inflammation is such an intrinsic component of RA.

Our objective was to conduct a meta-analysis of observational studies to determine the magnitude of the risk of CVD mortality, as well as cause-specific mortality from IHD and CVAs, in patients with RA compared with the general population.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. AUTHOR CONTRIBUTIONS
  8. REFERENCES

Search strategies.

Medline, EMBase, and Lilacs databases were searched from their inception (1966, 1980, and 1982, respectively) to July 2005 by an experienced librarian to find primary references and published reviews. The authors also searched reference lists from retrieved articles and searched for publications from scientists known for publishing in the field of mortality in patients with RA. The following search terms were used alone and in combination: rheumatoid arthritis, cardiovascular disease, cerebrovascular accident, transient ischemic attack, risk, risk factors, survival rate, mortality, comorbidity, causality, cause of death, heart death, sudden death, cohort study, case–control study, and longitudinal study.

We selected peer-reviewed articles (case–control and cohort studies) that met the following inclusion criteria: 1) prespecified RA definition, 2) clearly defined CVD outcome, including IHD and CVAs, and 3) reported age- and sex-adjusted SMRs and 95% confidence intervals (95% CIs) or data to calculate them. If data were duplicated in more than 1 study, the most recent study was included in the analysis.

Data extraction.

Two researchers (JAA-Z and DL) independently assessed studies for eligibility and extracted data on year of publication, type of study, source of RA population, RA definition, sample size, enrollment period, RA duration at cohort inception, mean time of followup, extent of loss of followup, reference group, outcome definition, number of observed and expected deaths from CVD, IHD, and CVAs for all, and by sex if available, matching or adjustment for cardiovascular risk factors, and ascertainment method for death. Where their initial conclusions did not agree, the researchers met to achieve consensus.

Quality scores of included studies.

We assessed study quality based on a 12-point scale that included elements of previously published scales for observational studies and adapted to the needs of the present meta-analysis (37, 38). Each study was scored according to 6 characteristics related to patients and methods of each study. Each item was scored as 0, 1, or 2. Specifically, we determined the source of study population (community-based [2], clinic-based [1], or undefined [0]), cohort type (inception [2], non-inception [1]), RA definition (use of current American College of Rheumatology [ACR; formerly the American Rheumatism Association] classification criteria for RA [2] [39], other validated criteria [1], other predefined but non-validated criteria [0]), ascertainment of CVD outcome (cause of death verified, e.g., medical records [2], cause of death not verified, e.g., International Classification of Diseases, Ninth Revision codes on death certificate [1], not mentioned [0]), extent of loss of followup (<20% [2], 20–40% [1], >40% [0]), and matching by or adjustment for Framingham risk factors (≥5 risk factors [2], <5 risk factors [1], none [0]). The complete protocol for quality scoring is available upon request from the corresponding author. For stratification purposes, studies that scored ≥10 were considered to be of higher quality and the rest were considered to be of lower quality. Quality scoring was performed independently by 2 reviewers (JAA-Z and DL). Disagreement was resolved by consensus.

Statistical analysis.

We calculated weighted–pooled summary estimates of SMRs (meta-SMRs) for all CVD, as well as for IHD and CVAs. The meta-SMR represents a summary estimate of the increased risk of death from CVD in patients with RA compared with the general population, weighted by the sample size of each study. Separate meta-SMRs were calculated for men and women when available. Calculations were performed on the log of the SMRs from the individual studies, and the resulting pooled values were then transformed back to the SMR scale. We used the random-effects model and tested for heterogeneity using the bootstrap version of the Q and I2 statistics using Stata statistical software, version 8.2 (40).

Because heterogeneity is expected in meta-analyses of observational studies, a subgroup analysis was carried out to assess heterogeneity. Studies were stratified based on study population (community-based versus clinic-based), cohort type (inception versus non-inception), enrollment period (before 1987 versus after 1987, to distinguish between current ACR classification criteria for RA [39] and older criteria), followup (≥10 years versus <10 years), and quality score (≥10 versus <10). Statistical inferences about the difference in the meta-SMRs between subgroups of studies were performed using a univariate meta–regression analysis (41). A multivariate meta–regression analysis evaluating the adjusted effect of the above study characteristics was not performed because of the small number of studies and because in the majority of studies, at least 1 study characteristic could not be estimated, leaving too few observations for the multivariate regression.

Robustness of the results was evaluated using a jackknife sensitivity analysis; i.e., the analysis was repeated multiple times, each time with removal of a single study from the baseline group of studies (42).

Assessment of publication bias/small-study effect.

To detect the presence of publication bias (i.e., the bias resulting from the greater likelihood of publication of studies reporting positive results compared with negative results) or the small-study effect (a tendency for treatment effect estimates in small studies to differ from those in larger studies) (43), we constructed a funnel plot, in which a measure of the study size is plotted as a function of the measure of interest (44). Again, we used the log of the SMRs from the individual studies as well as the log of precision (1/variance). If publication bias and small-study effect are absent, the distribution of the data points will be symmetric. In addition, Egger's regression was used to provide an objective, quantitative test statistic (P value) for the presence of asymmetry in the data (45).

RESULTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. AUTHOR CONTRIBUTIONS
  8. REFERENCES

We screened 578 abstracts published over the last 39 years. A total of 51 studies were retrieved for detailed evaluation and 24 studies were included (1–15, 19–27) (Figure 1). Twenty-seven studies out of 51 were excluded: 11 did not provide SMRs or data to calculate SMRs, 13 provided only all-cause mortality with no data to calculate CVD mortality, 2 were repeated studies on the same sample, and 2 were cross-sectional studies. The complete list of references reviewed is available upon request from the corresponding author.

thumbnail image

Figure 1. Meta-analysis of 24 studies on cardiovascular disease mortality in patients with rheumatoid arthritis.

Download figure to PowerPoint

The 24 studies included 111,758 patients with 22,927 cardiovascular events. All were cohort studies (Table 1). Eighteen of these studies were performed in Europe and 6 in North America. Eighteen of these studies used clinic-based population samples (n = 48,091), whereas 6 were community-based samples (n = 63,667).

Table 1. Characteristics of the 24 studies included in the meta-analysis of cardiovascular mortality in RA*
Author, yearCountryNo. patientsCVD eventsSettingEnrollment periodSubjects' mean age at entry, yearsFemale rateRA definitionMean followup, yearsOutcome ascertainmentQualityInception cohort
  • *

    RA = rheumatoid arthritis; CVD = cardiovascular disease; NA = not available; ARA = American Rheumatism Association; DC = death certificate; ICD = International Classification of Diseases; MR = medical record; ACR = American College of Rheumatology; ICD-7 = ICD, Seventh Revision; ICD-8 = ICD, Eighth Revision; ICD-9 = ICD, Ninth Revision; ICD-10 = International Statistical Classification of Diseases and Related Health Problems, Tenth Revision.

  • Per 100,000.

Uddin, 1970Canada47551Clinic-based1954–1966NA0.68ARANADC, autopsy7No
Monson, 1976US1,035311Clinic-based1930–1960NA0.74Clinical by specialist11.4DC6No
Lewis, 1980UK31115Clinic-based1966–1976NANAARA6.5DC, autopsy6No
Linos, 1980US521245Community-based1950–1974NA0.74ARANADC11Yes
Allebeck, 1982Sweden1,165144Clinic-based1971NA0.78ICD codes for RA7DC, autopsy6No
Vandenbroucke, 1984The Netherlands20971Clinic-based1954–1957540.66ARA25Family doctor9No
Mutru, 1985Finland1,000166Clinic-based1959–196855.30.5ARA10DC9No
Erhardt, 1989UK10720Clinic-based1976–1979590.66ARA∼6DC, MR8No
Reilly, 1990UK10018Clinic-based1957–196350.60.64ARA25DC, autopsy, MR10Yes
Jacobsson, 1993US2,97927Community-based (Aboriginals)1965–1989NANAARA and ACRNADC11No
Wolfe, 1994Canada and US3,501418Clinic- and community- basedCanada, 1966–1974; US, 1965–199053.30.74ARA and ACRCanada, 15.8; US, 8.5DC9No
Wallberg-Jonsson, 1997Sweden606140Clinic-basedUp to 1979NA0.68ARAFollowed until 1994DC11No
Symmons, 1998UK448104Clinic-based1964–197847.50.65ARA21.5DC9No
Turesson, 1999Sweden48963Clinic-based1990–199469.30.73ACR4.5DC9No
Kvalvik, 2000Norway14729Clinic-based1977580.65ARA15DC9No
Riise, 2001Norway1876Clinic-based1978–198259.90.69ARA12DC, MR10No
Bjornadal, 2002Sweden46,91712,431Community-based1964–1994NA0.71ICD-7, ICD-8, and ICD-910.4DC9No
Goodson, 2002UK57532Community-based1990–1994570.68ACR6.9DC10Yes
Thomas, 2003UK33,3187,185Clinic-based1981–200061.80.73ICD-9 and ICD-106.9DC7No
Watson, 2003UK11,633807Community-based1987 to last encounterNA0.70ICD codes5MR9No
Krishnan, 2004US3,862208Clinic-based1980–1997560.76ACR6.5DC9No
Sihvonen, 2004Finland1,042164Community-based1987NANAARA12DC, MR, autopsy11No
Book, 2005Sweden15252Clinic-based1978610.57ARA12.4DC9No
Goodson, 2005UK979220Clinic-based1981–199664.40.72Clinical by specialist11.5 (median)DC9Yes

There was a significantly increased mortality risk of CVD in patients with RA (meta-SMR 1.50, 95% CI 1.39–1.61) (Figure 1). Eight studies provided estimates by sex (3, 8, 11, 12, 15, 23, 24, 26). Overall, there was no clear difference between sexes (meta-SMR 1.58, 95% CI 1.35–1.84 for women; meta-SMR 1.45, 95% CI 1.11–1.90 for men).

We identified significant heterogeneity among studies (I2 = 0.93, P = 0.0001). Subgroup analyses showed that a number of factors influenced mortality risk (Table 2). Meta-SMRs were higher in studies with lower quality scores (<10), in studies with samples assembled after 1987, in non–inception cohorts, and in clinic-based samples. Inception cohort studies were the only group that did not show a significantly increased mortality risk for all CVD compared with the general population (meta-SMR 1.19, 95% CI 0.86–1.68), although the pooled sample size was small (n = 2,175). Despite the observed differences in mortality among subgroups, only quality scores were significantly associated with the observed heterogeneity in the meta–regression analysis (P = 0.02). However, a trend was also observed for cohort type (P = 0.09).

Table 2. Overall mortality and sensitivity analyses for the 24 cohort studies in patients with rheumatoid arthritis*
Study subsetNo. studiesNo. patientsNo. CVD eventsRandom-effects meta-SMR (95% CI)P
  • *

    CVD = cardiovascular disease; meta-SMR = weighted–pooled summary estimates of standardized mortality ratios; 95% CI = 95% confidence interval; NS = not significant.

  • Difference in the meta-SMRs between subgroups using univariate meta–regression analysis.

All studies24111,75822,9271.50 (1.39–1.61) 
Study population     
 Community-based663,66713,7061.35 (1.11–1.63)NS
 Clinic-based1848,0919,2211.53 (1.37–1.71) 
Cohort type     
 Inception42,1755151.19 (0.86–1.64)0.09
 Non-inception20109,58322,4121.56 (1.45–1.68) 
Quality score     
 Higher quality (≥10)76,0106311.21 (1.06–1.39)0.02
 Lower quality (<10)17105,74822,2961.57 (1.46–1.70) 
Enrollment period     
 Before 19871714,5501,9811.42 (1.22–1.66)NS
 After 1987797,19820,9461.67 (1.55–1.81) 
Followup length     
 Less than 10 years954,9618,8931.66 (1.45–1.90)NS
 More than 10 years1252,82213,7121.48 (1.32–1.66) 

Results of the jackknife sensitivity analysis are shown in Table 3. The meta-SMR remained significantly increased when studies were excluded 1 at a time, with the point estimates ranging from 1.41–1.54 and the corresponding 95% CIs remaining >1 in all cases.

Table 3. Sensitivity analysis using the jackknife approach, where each study is excluded at the time to test robustness of the overall SMR*
Author, yearSMR (95% CI) for CVDMeta-SMR (95% CI) when study excludedWeight (random effects)
  • *

    SMR = standardized mortality ratio; 95% CI = 95% confidence interval; CVD = cardiovascular disease; meta-SMR = weighted–pooled summary estimates of SMRs.

All studies1.50 (1.39–1.61)Not applicable 
Uddin, 19700.82 (0.64–1.01)1.54 (1.44–1.66)4.2
Monson, 19761.69 (1.51–1.89)1.49 (1.38–1.61)5.5
Lewis, 19801.43 (0.80–2.24)1.50 (1.39–1.62)1.9
Linos, 19801.09 (0.96–1.24)1.53 (1.43–1.65)5.2
Allebeck, 19821.45 (1.26–1.65)1.50 (1.39–1.62)5.2
Vandenbroucke, 19840.90 (0.70–1.2)1.53 (1.42–1.64)3.2
Mutru, 19851.39 (1.19–1.61)1.51 (1.40–1.62)5.0
Erhardt, 19892.38 (1.45–3.53)1.48 (1.38–1.60)2.3
Reilly, 19901.06 (0.63–1.58)1.51 (1.40–1.63)2.2
Jacobsson, 19931.77 (1.10–2.84)1.49 (1.39–1.61)1.8
Wolfe, 19942.30 (2.09–2.54)1.46 (1.36–1.58)5.6
Wallberg-Jonsson, 19971.46 (1.23–1.71)1.50 (1.39–1.62)4.9
Symmons, 19982.20 (1.80–2.64)1.47 (1.36–1.59)4.5
Turesson, 19991.75 (1.34–2.2)1.49 (1.38–1.61)3.9
Kvalvik, 20001.29 (0.85–1.81)1.51 (1.40–1.62)2.7
Riise, 20011.20 (0.90–1.8)1.51 (1.40–1.62)2.2
Bjornadal, 20021.81 (1.78–1.85)1.46 (1.33–1.63)6.2
Goodson, 20020.91 (0.62–1.26)1.52 (1.41–1.64)2.8
Thomas, 20031.93 (1.89–1.97)1.46 (1.33–1.61)6.2
Watson, 20031.50 (1.40–1.6)1.50 (1.39–1.62)6.0
Krishnan, 20041.59 (1.36–1.86)1.49 (1.38–1.61)4.9
Sihvonen, 20041.23 (1.05–1.43)1.41 (1.41–1.63)5.0
Book, 20051.57 (1.17–2.05)1.49 (1.38–1.61)3.4
Goodson, 20051.73 (1.51–1.97)1.49 (1.38–1.61)5.2

The funnel plot did not show a lack of small studies with negative results (Figure 2). Nevertheless, there seemed to be a difference in the effect size between small and large studies, possibly indicating the small-study effect (43). The Egger's test for asymmetry was also significant (P = 0.002). The observed asymmetry appeared to be mainly caused by the 2 studies with large sample sizes and strongly significant SMRs (10, 11). When these 2 studies were removed, the Egger's test was no longer significant (P = 0.34).

thumbnail image

Figure 2. Funnel plot of 24 studies evaluating mortality in patients with rheumatoid arthritis compared with the general population. Each dot represents individual studies. The solid line is the random-effects pooled estimate of log (standardized mortality ratio [SMR]).

Download figure to PowerPoint

There were 13 studies describing cause-specific CVD mortality, including mortality from IHD and CVAs (Figure 3). For IHD (n = 100,878 patients), 3 studies provided estimates by sex. Overall, there was an increased risk of death from IHD (meta-SMR 1.59, 95% CI 1.46–1.73), with no significant difference between sexes. Again, the jackknife sensitivity analysis demonstrated that the results were not influenced by any particular study, with the meta-SMR estimates ranging from 1.54–1.66 and the corresponding 95% CIs remaining >1.

thumbnail image

Figure 3. Meta-analysis of studies with cause-specific cardiovascular disease mortality in patients with rheumatoid arthritis.

Download figure to PowerPoint

Twelve studies provided information on mortality from CVAs (n = 100,285 patients) (Figure 3). Overall, there was an increased risk of death from CVAs (meta-SMR 1.52, 95% CI 1.40–1.67). There was no significant difference between sexes. The jackknife sensitivity analysis of the meta-SMR estimate showed that the pooled estimate was robust, with the point estimates varying from 1.54–1.66 and the corresponding 95% CIs remaining >1.

DISCUSSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. AUTHOR CONTRIBUTIONS
  8. REFERENCES

Our meta-analysis of published mortality studies of patients with RA indicates that there was a 50% increased risk of CVD mortality compared with the general population. We found no significant difference between sexes, and the increased mortality was attributable to increased deaths from IHD and CVAs.

The risk was higher in studies enrolling patients with RA after 1987 (meta-SMR 1.67, 95% CI 1.55–1.81) when compared with patients enrolled before 1987 (meta-SMR 1.42, 95% CI 1.22–1.66), suggesting that the use of current ACR classification criteria led to inclusion of cases with better defined RA, or possibly more severe disease.

As expected, samples recruited from clinics rather than community had a higher risk of CVD mortality. The only subgroup of studies that did not yield a significantly increased risk of mortality from CVD was inception cohorts (Table 2). This is likely due to the shorter duration of followup in these cohorts, and possibly to the smaller number of cases (n = 2,175). Of interest, the only inception cohort that showed a significantly increased risk of mortality from all CVD had a median followup of 11.4 years, and was also a clinic-based RA sample assembled after 1987 (15). This suggests that there might be a latent period after RA diagnosis until the risk of death from CVD is increased. On the contrary, we did not see a trend of increasing SMR with increasing of followup in the non–inception cohort studies (prevalent cases). We believe that inception cohort studies with longer followup are needed in order to reveal the best estimate for the risk of CVD death in patients with RA.

Similar to our findings, Ward found that study design accounted for most of the differences observed in survival rates in his review of 18 studies (46). However, Ward evaluated only all-cause mortality, but not CVD mortality. Unlike Ward's study, we pooled our data using a meta-analysis, whereas he averaged the estimates of individual studies to obtain a summary of SMRs.

Our study also demonstrates that risk of death from both IHD and CVAs is increased in patients with RA compared with the general population. The observed risk of death from CVAs was increased in some studies (1, 6, 10–12, 15, 25), but not in others (2, 4, 7, 9). Most studies where the risk was increased had large samples, suggesting that mortality from CVAs may be less frequent than from IHD, and lack of power may be the main reason for some studies not identifying an increased risk of death from CVAs.

Our study has some limitations. In our meta-analysis, we included cohorts that were clinically different in terms of age at enrollment, disease duration, disease severity, classification criteria to define RA, and study design. Heterogeneity in the results was observed as expected in meta-analyses of observational studies (44). As recommended for meta-analyses of observational studies, we used the random-effects model to include an estimate of the between-study variability (47). Interestingly, only quality score was able to explain some of the observed heterogeneity (P = 0.02), although cohort type showed a trend (P = 0.09) (Table 2). Furthermore, the small-study effect and/or publication bias might have compromised the validity of our results, and our estimate should be considered as tentative.

In this meta-analysis, the SMR evaluated the association between RA and CVD mortality adjusted for age and sex only. Although other confounding factors may influence the risk of CVD mortality in patients with RA, there is no method for adjusting the results of meta-analyses using SMRs. Nevertheless, although unadjusted confounders may influence the validity of meta-analyses of SMRs, several studies have shown that the increased risk of CVD in patients with RA is independent of traditional risk factors (48–50).

The majority of studies included in this meta-analysis enrolled patients before the widespread use of biologics; therefore, the results obtained may not be generalizable to RA samples treated with biologics. Accordingly, recent evidence suggests that all-cause and cause-specific mortality are not greater than expected in RA patients treated with biologic therapy (28). Therefore, more studies specifically evaluating mortality in RA patients treated with biologics would be valuable.

In summary, published data indicate that CVD mortality is increased by ∼50% in RA patients compared with the general population. However, the CVD mortality in the inception cohort subgroups was not elevated; therefore, our results might have been affected by an upward bias from non–inception cohort studies. Finally, the increased risk of death from all CVD, as well as death from IHD and CVAs, does not seem to differ between sexes; however, few studies reported sex-specific SMRs.

AUTHOR CONTRIBUTIONS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. AUTHOR CONTRIBUTIONS
  8. REFERENCES

Dr. Aviña-Zubieta had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study design. Aviña-Zubieta, Choi, Etminan, Esdaile, Lacaille.

Acquisition of data. Aviña-Zubieta, Choi, Lacaille.

Analysis and interpretation of data. Aviña-Zubieta, Choi, Sadatsafavi, Etminan, Lacaille.

Manuscript preparation. Aviña-Zubieta, Choi, Sadatsafavi, Etminan, Esdaile, Lacaille.

Statistical analysis. Aviña-Zubieta, Sadatsafavi, Lacaille.

Literature search. Kathy Hornby (nonauthor).

REFERENCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. AUTHOR CONTRIBUTIONS
  8. REFERENCES