Chapter  81:  Diagnosis and Treatment of Coronary Heart Disease in Women: Systematic Reviews of Evidence on Selected Topics

Overall

• Findings from 82 otherwise eligible studies could not be included in the systematic reviews because data were not stratified by gender. We contacted the author of these studies twice requesting data for women and received data from 19 studies (23 percent).

• Little evidence was available regarding the key questions as they pertain to women of different races/ethnicities. For this reason, only the review of diabetes as a risk factor for CHD provides summary findings by ethnicity.

Accuracy of exercise myocardial perfusion imaging and echocardiography for diagnosis of CHD in women

• We found 14 eligible studies that provided data on the accuracy of noninvasive tests in 893 women. Ten studies examined the accuracy of myocardial perfusion imaging and four examined the accuracy of exercise echocardiography.

• In women, the overall accuracy of both exercise myocardial perfusion imaging and exercise echocardiography for diagnosis of CHD is low with positive likelihood ratios of 2.5 to 3 and negative likelihood ratios of about 0.3.

• The accuracy of exercise myocardial perfusion imaging for diagnosis of CHD is not clinically different in women compared to men.

• There is little difference in the accuracy of exercise myocardial perfusion imaging and exercise echocardiography for diagnosis of CHD in women.

• The accuracy of exercise myocardial perfusion imaging for diagnosis of CHD is similar whether thallium or sestamibi is used as the imaging agent.

Efficacy of lipid lowering to reduce risk of CHD in women

• Although 20 clinical trials of the effects of lipid lowering therapy included women, only nine published results by gender. By contacting study investigators, we were successful in obtaining data on women from two additional trials. Thus, we were able to analyze results from 11 trials that included 15,917 women.

• In women with known CHD, treatment with lipid lowering therapy reduces risk of CHD mortality 26 percent, nonfatal myocardial infarction (MI) 36 percent and major CHD events 21 percent. There was insufficient evidence to show that lipid lowering reduces rates of revascularization procedures and no evidence of a reduction of risk in total mortality.

• For women without CHD, there is insufficient evidence to determine whether lipid lowering reduces risk for any clinical outcome.

Diabetes as a risk factor for CHD in women

• We found 17 eligible studies that included 43,944 women (4,522 with diabetes and 39,422 without diabetes).

• Adjusted summary odds ratios (ORs) for CHD mortality and nonfatal MI due to diabetes are higher among women than men, but summary ORs for all-cause mortality are slightly higher in men than women. All of the differences between men and women are modest and not statistically significant.

• The summary odds ratio for CHD mortality due to diabetes is 2.9 (95% confidence interval [CI], 2.2-3.8) for women and 2.3 (95% CI, 1.9-2.8) for men. The summary OR for nonfatal MI due to diabetes is 1.7 (95% CI, 1.3-2.3) for women and 1.6 (95% CI, 1.1-2.2) for men. The summary OR for all-cause mortality due to diabetes is 1.9 (95% CI, 1.7-2.3) for women and 2.1 (95% CI, 1.7-2.7) for men.

• Summary estimates for risk of CHD mortality due to diabetes for nonwhite men and women are similar to those for whites.

• The difference in relative risk for CHD outcomes between men and women is progressively attenuated with adjustment for major cardiovascular risk factors. This finding may be due to the fact that women with diabetes have more risk factors or more severe risk factor abnormalities in comparison to women without diabetes than is the case for men with and without diabetes.

Prognostic value of troponin for CHD in women

• We identified eight eligible cohort studies that provided data on 3,169 women and 4,070 men.

• Elevated troponin was observed in 35 percent of women and 39 percent of men with non-ST elevation acute coronary syndromes.

• Women with acute coronary syndromes were older and more likely to have diabetes and hypertension than men.

• An elevated troponin indicates a similar increase in risk of death for both women (summary OR 2.63; 95% CI 1.75-3.95) and men (OR 2.83; 95% CI 1.92-4.17).

• An elevated troponin indicates a greater increase in risk of nonfatal MI for women (summary OR 1.80; 95% CI 1.28-2.54) than men (OR 1.06; 95% CI 0.8-1.41).

Conclusions. The major problem in performing these systematic reviews was that data stratified by sex and race/ethnicity from completed studies are often not available. We recommend that, in addition to requiring participation of women and minorities in research, the National Institutes of Health, U.S. Food and Drug Administration, and other funding and regulatory agencies insist that outcome data by subgroup be published or archived and made easily available to meta-analysts.

Accuracy of exercise myocardial perfusion imaging and echocardiography for diagnosis of CHD in women

We searched PubMed^®, the Cochrane Database, and DARE for articles in English and other languages published from 1990 through January 2002. We used the following search terms to identify cross-sectional studies in which the accuracy of the exercise MPI or exercise echocardiography was compared to angiographic findings:

( Note: An asterisk indicates truncation of the search term.)

• Exercise MPI: thallium radioisotopes, radiopharmaceuticals, tomography emission-computed single-photon, technetium TC 99M sestamibi, organotechnetium compounds, Spect, Cardiolite, Mibi

  AND exercise, exercise test, exercise tolerance, exercise*, exercising, "stress test" AND diagnosis, diagnoses, diagnostic, diagnosing, predictive values of test

• Exercise echocardiography: echocardio*, ultrasound, ultrasonography

  AND exercise, exercise test, exercise tolerance, exercise*, exercising, "stress test"

  AND diagnosis, diagnoses, diagnostic, diagnosing, predictive values of test

• Outcomes: cardiovascular diseases, heart diseases, myocardial ischemia,

  coronary disease

Searches for noninvasive diagnostic tests identified 3,136 titles. After eliminating ineligible studies by review of titles and abstracts, we reviewed the full text of 326 articles and found 14 eligible cross-sectional studies with data on women that were included in the systematic review. Ten studies examined the accuracy of MPI and four examined the accuracy of exercise echocardiography.

Efficacy of lipid lowering to reduce risk of CHD in women

We searched PubMed^®, the Cochrane Database, and DARE for articles in English and other languages published from 1966 through January 2002. We used the following search terms to identify clinical trials:

• Lipid lowering: hyperlipidemia and anticholesteremic agents, antilipemic agents, simvastatin, lovastatin, pravastatin, atorvastatin, fluvastatin, gemfibrozil, cholestyramine, cholestpol, niacin

• Outcomes: cardiovascular diseases, heart diseases, myocardial ischemia, coronary disease

Searches for clinical trials of lipid lowering treatment identified 1,335 titles. After eliminating ineligible studies by review of titles and abstracts, we reviewed the full text of 120 articles and found 11 eligible randomized trials that provided data on women and were included in the systematic review.

Diabetes as a risk factor for CHD in women

We searched PubMed^®, the Cochrane Database, and DARE for articles in English and other languages published from 1966 through January 2002. We used the following search terms to identify cohort and cross-sectional studies:

• Diabetes: diabetes

• Outcomes: cardiovascular disease, myocardial infarction, ischemic heart disease

Searches for diabetes as a risk factor for CHD in women identified 4,578 titles. After eliminating ineligible studies by review of titles and abstracts, we reviewed the full text of 233 articles. We found 17 studies that fulfilled all inclusion criteria; 12 were prospective cohort studies and five were cross-sectional analyses.

Prognostic value of troponin for CHD in women

We searched MEDLINE^® for articles in English and other languages published from 1966 through January 2002. We used the following search terms to identify clinical trials or cohort studies:

• The text word troponin, and

• The text words angina or unstable or myocardial infarction or ischemia.

We also performed a search of EMBASE from 1990-1998, but did not find any additional articles fulfilling the study criteria.

Searches identified 1,049 articles. We excluded 878 articles based on title or abstracts and reviewed the full text of 171 articles. Of these, eight eligible studies provided data on women and were included in the systematic review; six were clinical trials and two were cohort studies.

Overall

• Data from many otherwise eligible studies could not be included in the systematic reviews because the findings were not stratified by sex. We identified 82 studies that included women, but did not stratify the data by sex. We contacted authors of these studies twice requesting data on women but received data from only 19 studies (23 percent).

• Little evidence was available regarding the key questions as they pertain to women of different races/ethnicities. For this reason, only the review of diabetes as a risk factor for CHD provides summary findings by ethnicity.

Accuracy of exercise myocardial perfusion imaging and echocardiography for diagnosis of CHD in women

• Although 34 eligible studies of the accuracy of exercise myocardial perfusion imaging or exercise echocardiography included women, only nine published results by sex. By contacting study investigators, we were successful in obtaining data on women from five additional studies. Thus, we were able to analyze results from 14 studies that included 893 women. Ten studies examined the accuracy of myocardial perfusion imaging and four examined the accuracy of exercise echocardiography.

• In women, the overall accuracy of both exercise myocardial perfusion imaging and exercise echocardiography for diagnosis of CHD is low with positive likelihood ratios of 2.5 to 3 and negative likelihood ratios of about 0.3.

• The accuracy of exercise myocardial perfusion imaging for diagnosis of CHD is not clinically different in women compared to men.

• There is little difference in the accuracy of exercise myocardial perfusion imaging and exercise echocardiography for diagnosis of CHD in women.

• The accuracy of exercise myocardial perfusion imaging for diagnosis of CHD is similar whether thallium or sestamibi is used as the imaging agent.

Efficacy of lipid lowering to reduce risk of CHD in women

• Although 20 clinical trials of the effects of lipid lowering therapy included women, only nine published results by sex. By contacting study investigators, we were successful in obtaining data on women from two additional trials. Thus, we were able to analyze results from 11 trials that included 15,917 women.

• In women with known CHD, treatment with lipid lowering therapy reduces risk of CHD mortality 26 percent, nonfatal myocardial infarction (MI) 36 percent and major CHD events 21 percent. There was insufficient evidence to show that lipid lowering reduces rates of revascularization procedures and no evidence of a reduction of risk in total mortality.

• For women without CHD, there is insufficient evidence to determine whether lipid lowering reduces risk for any clinical outcome.

Diabetes as a risk factor for CHD in women

• Although 36 eligible studies included women, only 10 published results by sex. By contacting study investigators, we were successful in obtaining data on women from seven additional studies. Thus, we were able to analyze results from 17 studies that included 43,944 women (4,522 with diabetes and 39,422 without diabetes).

• Adjusted summary odds ratios (ORs) for CHD mortality and nonfatal MI due to diabetes are higher among women than men, but summary ORs for all-cause mortality are slightly higher in men than women. All of the differences are modest and not statistically significant.

• The summary OR for CHD mortality due to diabetes is 2.9 (95% confidence interval [CI], 2.2-3.8) for women and 2.3 (95% CI, 1.9-2.8) for men. The summary OR for nonfatal MI due to diabetes is 1.7 (95% CI, 1.3-2.3) for women and 1.6 (95% CI, 1.1-2.2) for men. The summary OR for all-cause mortality due to diabetes is 1.9 (95% CI, 1.7-2.3) for women and 2.1 (95% CI, 1.7-2.7) for men.

• Summary estimates for risk of CHD mortality due to diabetes for white men and women are similar to those for all ethnicities combined.

• The difference in relative risk for CHD outcomes between men and women is progressively attenuated with adjustment for major cardiovascular risk factors. This finding may be due to the fact that women with diabetes have more risk factors or more severe risk factor abnormalities in comparison to women without diabetes than is the case for men with and without diabetes.

Prognostic value of troponin for CHD in women

• We reviewed the full text of 171 articles and found three eligible studies with data on women. Nine additional large studies of the prognostic value of troponin included women, but did not provide data stratified by sex. After contacting authors, we obtained data for women from five of these studies. Thus, we identified eight eligible studies that provided data on 3,169 women and 4,070 men.

• Elevated troponin was observed in 35 percent of women and 39 percent of men with non-ST elevation acute coronary syndromes.

• Women with acute coronary syndromes were older and more likely to have diabetes and hypertension than men with acute coronary syndromes.

• Elevated troponin indicates a similar increase in risk of death for both women (summary OR 2.63; 95% CI, 1.75-3.95) and men (summary OR 2.83; 95% CI, 1.92-4.17).

• Elevated troponin indicates a greater increase in risk of nonfatal MI for women (summary OR 1.80; 95% CI, 1.28-2.54) than men (summary OR 1.06; 95% CI, 0.8-1.41).

Overall

• Future studies that include women should publish or make available outcomes stratified by sex and ethnicity.

Accuracy of exercise myocardial perfusion imaging and echocardiography for diagnosis of CHD in women

• The quality of future studies of the accuracy of noninvasive tests for the diagnosis of CHD should be improved by excluding persons with known CHD, performing both the noninvasive test and angiography in all participants and assuring that the outcome of the noninvasive test is assessed by personnel blinded to the results of angiography.

• Future research should address ways to improve accuracy of noninvasive tests for CHD in both men and women.

Efficacy of lipid lowering to reduce risk of CHD in women

• Future clinical trials should include adequate numbers of women to determine the effect of lipid lowering in women at high risk but without known CHD.

Diabetes as a risk factor for CHD in women

• Future prospective studies should present sex- and race/ethnicity-specific fatal and nonfatal coronary disease endpoints before and after adjustment for established CHD risk factors.

• Future studies should attempt to clarify the effect of established risk factors, which cluster in women with diabetes, compared to the effect of diabetes itself in increasing risk for CHD among women with diabetes.

Prognostic value of troponin for CHD in women

• Future studies are needed to verify and explore why the prognostic value of elevated troponin results for nonfatal MI is different in women compared to men.

Data sources

We searched PubMed^®, the Cochrane Database, and DARE for articles in English and other languages published from 1990 through January 2002. We also reviewed bibliographies and asked peer reviewers (Appendix A) to identify additional articles. The date limits of the search were chosen because both exercise echocardiography and exercise MPI using SPECT with thallium and sestamibi were in widespread use during this period.

Search Terms

We used the following search terms to identify cross-sectional studies in which the accuracy of the diagnostic tests of interest were compared to angiographic findings:

Inclusion Criteria

To be included, articles were required to fit the following criteria:

• 1

  Contained primary data on at least 10 women who underwent exercise ECG with radionuclide injection and SPECT imaging or exercise echocardiography.

• 2

  Estimated accuracy of noninvasive tests using angiographic evidence of CHD as the gold standard.

• 3

  Provided data to calculate true positives (TP), true negatives (TN), false positives (FP) and false negatives (FN) for the noninvasive tests.

• 4

  Clear definition of positive noninvasive test and positive angiogram provided.

• 5

  Published between 1990 and January 2002. Articles published outside this date range that were recommended by peer reviewers were included.

We excluded studies that met the following criteria:

• 1

  Noninvasive tests performed exclusively in patients after myocardial infarction (MI), percutaneous angioplasty, coronary artery bypass surgery or hospitalization for an unstable coronary syndrome. In these patients, noninvasive tests are done for the purpose of risk assessment rather than diagnosis.

• 2

  Tests in which pharmacologic agents rather than exercise were used as the stressor. Use of pharmacologic stressors may significantly affect the accuracy of noninvasive testing; many different agents are used and protocols for their use vary substantially.

Article Identification

An initial search using the terms listed above identified articles that potentially provided evidence. Two University of California, San Francisco (UCSF)-Stanford Evidence-based Practice Center (EPC) investigators reviewed the titles and excluded those that clearly did not provide data on humans or clearly did not address the question.

The abstracts of the remaining articles were reviewed independently by two UCSF-Stanford EPC physician investigators and coded using the categories listed below. Disagreements were discussed and consensus codes were entered into a database (Access, Microsoft Corporation).

T Test – the study clearly does not include data on exercise ECG with imaging or exercise echocardiography.

A Angiogram - the study clearly does not compare the results of the noninvasive test with the results of angiography.

ND Not diagnostic - The study assesses noninvasive tests performed exclusively in patients after myocardial infarction, percutaneous angioplasty, coronary artery bypass surgery or hospitalization for an unstable coronary syndrome.

R Review – the study is a review that does not contain primary data.

NH No humans - the study clearly does not include data on humans.

E₁ Eligible – the study may contain primary evidence regarding the research questions in women and will be reviewed in full-text.

Articles coded E₁ were retrieved and the full text was reviewed independently by two UCSF-Stanford EPC physician investigators. Names of authors and titles of journals were obscured before articles were reviewed.

Obtaining Unpublished Results in Women

Some eligible studies included women in the study population, but did not report findings separately by gender. In these instances we attempted to contact authors of these studies to obtain estimates in women. If we did not receive a response after the first contact, a second attempt was made. We contacted 34 authors^{2, 9- 41} and received data from

five.^{11, 15, 17, 18, 24}

Quality Assessment

The full text of each eligible study was reviewed independently by two UCSF-Stanford EPC physician investigators who completed a quality evaluation form (Appendix B). The studies included in this systematic review are cross-sectional. The three major quality issues affecting these studies are verification bias, biased outcome measurement and spectrum effect. Verification bias occurs when the decision to proceed to the gold standard is in part dependent on the results of the noninvasive test. Since positive noninvasive test results are more likely to be followed by an invasive test, this tends to increase the chance of detecting a true positive (TP) relative to a false negative (FN) and tends to increase the chance of detecting a false positive (FP) relative to a true negative (TN). Therefore, sensitivity may appear to be higher and specificity lower in the verified sample. Biased outcome measurement occurs when personnel performing or reading the results of the noninvasive test already know the results of angiography. Spectrum effect refers to the variation in test performance depending on the severity of disease in the population studied. Sensitivity and specificity appear higher when the persons studied either have severe disease or are healthy. For instance, in participants with significant coronary disease and healthy volunteers, the spectrum of disease is clear-cut, and both sensitivity and specificity will be higher compared to a population with intermediate prior probability of coronary disease, such as those with angina. Our quality assessment addresses verification bias and biased outcome measurement, and we recorded spectrum of disease to allow subgroup analyses.

To be categorized as good quality, articles were required to meet the following parameters:

• All participants who had the noninvasive test also had angiography.

• The diagnosis of coronary artery disease on angiography was made by investigators blinded to the results of the noninvasive test

Studies that did not meet these criteria were considered fair quality.

Data Abstraction

Two UCSF-Stanford EPC physician investigators independently reviewed the full text of each eligible study and completed a data abstraction form (Appendix C). Data abstracted included characteristics of the study (design, inclusion and exclusion criteria, noninvasive tests performed, and setting), participant characteristics (number of women and men, mean age of participants, number with prior MI, number with revascularization, cardiac risk factors in the population, and indications for cardiac testing), and test characteristics (type of exercise, average duration of exercise, percent with adequate exercise, radionuclide and imaging protocols used, criteria for positive noninvasive test, and criteria for positive coronary angiogram). For each eligible study, the numbers of true positive, true negative, false positive and false negative tests were recorded or calculated as necessary. We also abstracted accuracy measures for all subgroups evaluated. Disagreements between abstractors were discussed and decided by consensus. For studies with multiple publications, only data from the most comprehensive or recent publication were used.

Data Management and Archive

We entered all identified titles and abstracts in an EndNote^® file (Niles Software, Inc.) that includes searchable key words as codes for eligibility. Information on all articles that were reviewed in full text was transferred from EndNote^® to a database (Access, Microsoft^® Corporation) that allows us to categorize each article by reason for exclusion. Quality assessment data for each eligible study were also entered in the database, allowing us to categorize eligible articles by quality.

Abstracted data were entered into a database (EXCEL, Microsoft^® Corporation) for preparation of evidence tables and calculation of summary estimates, confidence intervals and tests of heterogeneity.

The full-text articles that were retrieved, and the abstraction forms for each article are filed in Dr. Grady's offices at the UCSF Mt. Zion Women's Health Clinical Research Center.

Data Analyses

The primary outcomes of each study were expressed as sensitivity, specificity, positive likelihood ratio and negative likelihood ratio comparing the results of the noninvasive test to angiographic findings. Summary results were calculated as the mean of the appropriate proportion (sensitivity, specificity, likelihood ratios) weighted by the sample size of each individual study. The significance level for all p-values for the weighted means was set at 0.05. All findings were assessed for heterogeneity using Z-tests. The significance level for tests of heterogeneity was 0.10. To avoid calculation problems associated with zero cells, 0.5 was added to all cells to calculate variances and standard deviations.⁴² Results for women vs. men were compared using the Q* statistic, the point on the summary ROC curve where sensitivity equals specificity.⁴³

Publication bias usually occurs if small studies with unremarkable findings (poor accuracy) are not published while small studies with markedly positive findings (high accuracy) are published. We calculated the correlation between individual study sample size and sensitivity using Kendall’s Tau to assess potential publication bias.

Study Identification

Our searches identified 3,136 titles. After eliminating ineligible studies by review of titles and abstracts, we reviewed the full text of 326 articles and found 14 eligible for inclusion in the systematic review.^{2, 6, 17, 18, 23, 24, 41, 44- 50} Of the 10 studies included that examined exercise MPI as the noninvasive test, five used thallium,^{17, 41, 44- 46} four used sestamibi^{18, 23, 47, 48} and one used both⁶ as radionuclide agents.

Nine MPI studies that provided accuracy estimates in women were excluded from the systematic review: four used pharmacologic agents in addition to exercise;^{51- 54} four reported some measures of accuracy, but did not report adequate data to allow calculation of all required estimates,^{5, 55- 57} and one did not provide definitions of a positive noninvasive test or positive coronary angiogram.⁵⁸

Four eligible studies examined the accuracy of exercise echocardiography as the noninvasive test.^{2, 24, 49, 50} One study of exercise echocardiography that provided accuracy estimates for women was excluded because it was published outside our date range. ⁵⁹

Description of Eligible Studies

The characteristics of the 10 studies assessing the accuracy of MPI that were included in the systematic review are shown in Evidence Table 1. The number of participants in each study ranged from 41 to 371 and 14 to 100 percent were women. The total number of participants was 1,249, including 549 women and 700 men. Three of the studies included only women. One study included both men and women, but provided data that allowed calculation of accuracy estimates only in women.⁴⁸ The mean age of participants ranged from 51 to 62 years. Eight of the studies included participants with prior MI. All 10 studies used SPECT imaging; five used thallium only, four sestamibi only and one used both radionuclides. The definition of an abnormal test was very similar in nine of the studies (fixed or reversible perfusion defects, perfusion defects at rest or after exercise or decreased uptake at rest or after exercise). One study defined a positive test as reversible uptake defects in more than one of 22 coronary segments. All but one of the 10 studies defined 50 percent stenosis of one or more major coronary artery at angiography as the gold standard for the presence of coronary artery disease. Eight studies used treadmill exercise and two used bicycle exercise. Six of the MPI studies were judged fair quality and four were judged good quality.

The characteristics of the four studies assessing the accuracy of echocardiography that were included in the systematic review are shown in Evidence Table 2. The number of participants in each study ranged from 70 to 340 and 15 to 100 percent were women. The total number of participants was 689, including 344 women and 345 men. Two of the studies included both men and women and 2 included only women.^{2, 49} The mean age of participants ranged from 55 to 66 years. One of the studies included participants with prior MI. Two of the studies defined an abnormal test as new or worse regional wall motion abnormalities after exercise and two defined a positive as regional wall motion abnormalities at rest or after exercise. All studies defined 50 percent stenosis of at least one major coronary artery at angiography as the gold standard for the presence of coronary artery disease. Two of the echocardiography studies were judged fair quality and two were judged good quality.

Exercise Myocardial Perfusion Imaging

In women, sensitivity of exercise MPI using either thallium or sestamibi ranged from 0.61 to 1.0 with a mean weighted sensitivity of 0.77 (95% CI 0.72-0.83) (Evidence Table 3). Specificity of exercise MPI in women ranged from 0.40-1.0, with a mean weighted specificity of 0.69 (95% CI 0.62-0.75). The mean weighted positive likelihood ratio for exercise MPI in women was 2.46 (95% CI 2.00-3.04) and the mean weighted negative likelihood ratio was 0.33 (95% CI 0.26-0.41).

Based on the findings of the six studies that included men, the mean weighted sensitivity of exercise MPI in men was 0.93 (95% CI 0.90-0.95) (Evidence Table 3) and the mean weighted specificity was 0.57 (95% CI 0.47-0.67). The mean weighted positive likelihood ratio for exercise MPI in men was 2.17 (95% CI 1.73-2.73) and the mean weighted negative likelihood ratio was 0.13 (95% CI 0.09-0.19).

We performed a subgroup analysis limited to the findings of the four good quality studies (Evidence Table 1). Mean weighted accuracy estimates from these analyses were not materially different from the overall mean summary estimates (Evidence Table 3).

We performed two sensitivity analyses. One eligible study evaluated the accuracy of both thallium and sestamibi.⁶ The overall mean weighted results included only the accuracy estimates for sestamibi. A sensitivity analysis substituting the results for thallium produced similar overall results. We also repeated the analysis for women including the findings of one study that was excluded because no definitions of an abnormal test or abnormal angiogram were provided.⁵⁸ Including the results of this study did not materially change the accuracy estimates.

We calculated mean weighted accuracy estimates for men and women from studies that used sestamibi separately from those that used thallium (Evidence Table 4). Accuracy estimates in women for studies using sestamibi and those using thallium were very similar (p-value for the comparison of Q* statistics = 0.84)

Exercise Echocardiography

In women, sensitivity for exercise echocardiography ranged from 0.77 to 0.88 with a mean weighted sensitivity of 0.81 (95% CI 0.74-0.87) (Evidence Table 5). Specificity for exercise echocardiography in women ranged from 0.37 to 0.84, with a mean weighted specificity of 0.73 (95% CI 0.66-0.79). The mean weighted positive likelihood ratio for exercise echocardiography in women was 2.95 (95% CI 2.28-3.79) and the mean weighted negative likelihood ratio was 0.26 (95% CI 0.19-0.36).

Based on the findings of two studies of the accuracy of echocardiography that included men, the mean weighted sensitivity was 0.84 (95% CI 0.79-0.88) and the mean weighted specificity was 0.45 (95% CI 0.32-0.59) (Evidence Table 5). The mean weighted positive likelihood ratio for exercise echocardiography in men was 1.54 (95% CI 1.20-1.98) and the mean weighted negative likelihood ratio was 0.36 (95% CI 0.24-0.53).

We performed subgroup analyses limited to the findings of the two good quality studies of the accuracy of exercise echocardiography in women (Evidence Table 1). The mean weighted sensitivity estimate from the good quality studies was similar to the overall mean sensitivity (0.82 vs. 0.81), but specificity was lower (0.60 vs. 0.73) (Evidence Table 5). These differences resulted in a lower mean weighted positive likelihood ratio based on the good quality studies compared to the estimate based on all eligible studies (2.06 vs. 2.95). Mean estimates of negative likelihood ratios did not differ when results were restricted to the good quality studies (Evidence Table 5).

We performed a sensitivity analysis by adding the results of one study of the accuracy of exercise echocardiography that was published before our date range.⁵⁹ Including the results of this study did not materially change the overall accuracy estimates.

Accuracy of Exercise MPI vs. Echocardiography in Women

Based on 10 studies of the accuracy of exercise MPI and four of exercise echocardiography, the accuracy of each test for the diagnosis of CHD in women is similar (Evidence Tables 3 and 5). The mean weighted sensitivity, specificity and positive likelihood ratio for MPI are slightly lower than for echocardiography (sensitivity 0.77 vs. 0.81; specificity 0.69 vs. 0.73, positive likelihood ratio 2.46 vs. 2.95), but the differences are small and not statistically different (p-value for the Q* statistic = 0.10). The accuracy of the two tests is also similar when analyses are restricted to good quality studies (Evidence Tables 3 and 5). The use of sestamibi instead of thallium also did not change the accuracy of MPI studies in women (Evidence Table 4).

Accuracy of Noninvasive Testing in Women Compared to Men

The mean weighted sensitivity of MPI in women is somewhat lower than in men (0.77 vs. 0.93), but the specificity is higher (0.69 vs. 0.57) (Evidence Table 3). The positive likelihood ratio is slightly higher in women compared to men (2.46 vs. 2.17) as is the negative likelihood ratio (0.33 vs. 0.13). These differences in the accuracy of MPI between men and women were statistically significant (p-value for the comparison of Q* statistics 0.028), but it is not clear whether higher sensitivity or higher specificity is preferable.

Comparison of mean weighted accuracy estimates between women and men may be biased if these data are derived from different studies that may have used somewhat different methods and definitions of positive tests. To avoid this problem, we calculated mean weighted accuracy estimates for men and women restricted to the findings of studies that included both genders (Evidence Table 3). Based on the findings of these studies, sensitivity of MPI is lower in women than in men (0.86 vs. 0.93), but specificity is the same (0.57 for both genders; p-value for the comparison of Q* statistics 0.012). This analysis suggests that exercise MPI is more accurate in men than in women, but the differences are small and not clinically meaningful.

The mean weighted sensitivity of echocardiography in women is similar to that in men (0.81 vs. 0.84), but the specificity is higher (0.73 vs. 0.45) (Evidence Table 5). The positive likelihood ratio was substantially higher in women than in men (2.95 vs. 1.54), and the negative likelihood ratio was slightly lower (0.26 vs. 0.36). However, given the small numbers of men included in the analyses, we could not calculate a Q* statistic or determine any statistically significant differences between men and women with regard to exercise echocardiography.

Assessments for Heterogeneity and Publication Bias

There was no heterogeneity in any of the mean weighted estimates of accuracy. Publication bias usually occurs if small studies with unremarkable findings (poor accuracy) are not published while small studies with markedly positive findings (high accuracy) are published. We calculated the correlation between individual study sample size and sensitivity using Kendall’s Tau to assess potential publication bias. There was no evidence of publication bias in any of the summary estimates of accuracy.

Data sources

We searched PubMed^®, the Cochrane Database, and DARE for articles published in English and other languages from 1966 through January 2002. We also reviewed bibliographies and asked peer reviewers (Appendix A) to identify additional articles.

Search Terms

Search terms were developed in collaboration with a medical librarian and included the following:

Inclusion Criteria

To be included, articles were required to fit the following criteria:

• 1

  Randomized clinical trials of outpatients with or without known CHD.

• 2

  Treatment duration of at least one year.

• 3

  Study population classified as either primary (participants without prior CHD) or secondary prevention (participants with prior CHD).

• 4

  Data on women provided.

• 5

  Impact of lipid lowering with drug treatment assessed for at least one of the following clinical outcomes: total mortality, CHD mortality, nonfatal MI, CHD events or revascularization procedures. Coronary events included ischemic coronary syndromes and nonfatal myocardial infarction. CHD procedures included coronary bypass graft surgery and percutaneous coronary angioplasty or stenting.

• 6

  Published between January 1, 1966 and January 30, 2002. Articles published outside this date range recommended by peer reviewers were included.

We excluded studies that only provided evidence on the effect of treatment on changes in lipids, angiographic findings or other intermediate outcomes. For studies with multiple publications, only data from the most comprehensive or most recent publication were used.

Article Identification

An initial search using the terms listed above identified articles that potentially provided evidence. Two University of California, San Francisco (UCSF)-Stanford Evidence-based Practice Center (EPC) physician investigators reviewed the titles and excluded those that clearly did not provide data on humans or clearly did not address the question.

The abstracts of the remaining articles were reviewed independently by two UCSF-Stanford EPC physician investigators and coded using the categories listed below. Disagreements were discussed and the following consensus codes were entered into a database (Access, Microsoft Corporation):

RQ Research question - the article clearly does not address the research question.

R Review – the study is a review that does not contain primary data.

NSD Not appropriate study design - The article is not a randomized clinical trial.

NH No humans - the study clearly does not include data on humans.

E₁ Eligible – the study may contain primary evidence regarding the research question in women and will be reviewed in full-text.

Articles coded E₁ were retrieved and the full text was reviewed independently by two UCSF-Stanford EPC investigators. Names of authors and titles of journals were obscured before articles were reviewed.

Obtaining Unpublished Results

Some eligible studies included women in the study population, but did not report findings separately by gender. In these instances we attempted to contact authors to obtain these data. If we did not receive a response after the first contact, a second attempt was made. We contacted 13 authors^{11, 14- 27} and received data from two.^{16, 25, 26}

Quality Assessment

The full text of each eligible study was reviewed independently by two UCSF-Stanford physician investigators, who completed a quality evaluation form (Appendix B). All of the studies included in this systematic review are randomized clinical trials. To be categorized as good quality, articles were required to meet the following additional parameters:

• Inclusion/exclusion criteria clear and appropriate.

• Randomization allocation concealed.

• Control group received placebo.

• Participants and research staff blinded to intervention.

• More than 75 percent complete followup.

All other trials were considered fair quality. Disagreements between reviewers regarding quality parameters were decided by discussion and consensus.

Data Abstraction

Two UCSF-Stanford EPC physician investigators independently reviewed the full text of each eligible study and completed a data abstraction form (Appendix C). We abstracted information on the study population (primary prevention trials were defined as those that included individuals without known CHD; secondary prevention trials included individuals with known CHD), inclusion criteria, length of followup, numbers of men and women, participant characteristics such as age, other cardiovascular risk factors and cardiac medication use, baseline and followup lipoprotein values and all clinical outcomes that were measured. When possible, data were abstracted for men and women separately. Disagreements were discussed and decided by consensus.

Data Management and Archive

We entered all identified titles and abstracts in an EndNote^® file (Niles Software, Inc.) that includes searchable key words as codes for eligibility. Information on all articles that were reviewed in full text was transferred from EndNote^® to a relational database (Access, Microsoft^® Corporation) that allows us to categorize each article by reason for exclusion. Quality assessment data for each eligible study were also entered in the Access database, allowing us to categorize eligible articles by quality. Selected data were transferred to a flatfile database (EXCEL, Microsoft^® Corporation) for preparation of evidence tables and calculation of summary estimates, confidence intervals and tests of heterogeneity.

The full-text articles that were retrieved, and the abstraction forms for each article are filed by topic and question in Dr. Grady's offices at the UCSF Mt. Zion Women's Health Clinical Research Center.

Data Analyses

The primary outcome of each clinical trial was expressed as the relative risk (RR) among treated compared to untreated study participants. Summary estimates of RR and 95 percent confidence intervals (CI) were calculated using the Mantel-Haenszel method and both fixed and random effects models. Results of the fixed and random effects models were similar, and we report only the findings of the random effects model. To avoid calculation problems associated with zero cells, 0.5 was added to all cells to calculate variances and standard deviations.²⁸ The significance level for all tests of outcome was set at 0.05. All findings were assessed for heterogeneity using a standard Chi-square test and Q statistic with critical value set at 0.10. All analyses were performed separately for the findings of primary and secondary prevention studies. Subgroup analyses were performed by type of drug treatment (statins vs. others), and by good vs. fair quality.

Publication bias usually occurs if small studies with unremarkable findings (relative risks about 1.0) are not published while small studies with markedly positive findings (in this case, low relative risks) are published. We calculated the correlation between individual study weight (1/variance) and relative risk using a nonparametric correlation coefficient (Kendall’s Tau) with critical value set at 0.10 to assess potential publication bias. Statistically significant correlation of study weight and relative risk suggests publication bias.

Results of Study Identification

Our searches identified 1,335 titles. After eliminating ineligible studies by review of titles and abstracts, we reviewed the full text of 120 articles. We identified 20 studies that fit all inclusion criteria, but only 9 provided outcomes stratified by sex.^{7- 9, 11, 27, 29- 39} We contacted the principal investigators of the studies that did not provide data for women to request this information.^{11, 14- 27} We received data on women from two investigators.^{16, 25, 26} Thus, 11 studies (represented by 19 articles) were found to be both eligible and to contain data stratified by sex for inclusion in the systematic review.^{7- 9, 11, 16, 25- 27, 29- 39} One additional study did not meet inclusion criteria because it did not provide data on any of the clinical outcomes of interest and the study population was equally divided between persons with prior CHD and those without prior CHD and separate estimates for the effects of lipid lowering in primary and secondary prevention were not published.⁴⁰

Description of Eligible Studies

Characteristics of the 11 eligible trials are described in Evidence Table 6. The numbers of participants in each trial ranged from 151 to 20,536 and 15 to 50 percent of participants were women. The total number of women included in the trials was 15,917, but almost two thirds were from two studies.^{11, 39}Information on the ethnicity of participants was not provided in most trials. Duration of treatment ranged from 2.8 to 6.1 years and averaged 4.7 years. Seven of the trials were classified as secondary prevention and four were classified as primary prevention (Evidence Table 6). Eligibility criteria for 5 trials required at least mild hyperlipidemia,^{7, 16, 25, 26, 31, 39} four required a range of cholesterol levels that would include some participants in the normal range^{9, 11, 27, 34} and 2 included participants regardless of cholesterol levels.^{29, 30} Three trials assessed the effects of clofibrate^{29, 30} or colestipol³¹ and eight assessed the efficacy of treatment with statins (lovastatin,^{9, 25, 26} simvastatin,^{7, 11, 39}pravastatin^{16, 27, 34}).

All but one of the 11 trials included a placebo control group³⁹ and all but two were adequately blinded.^{31, 39} In all but one of the trials,³¹ followup was greater than 75 percent complete. Overall, seven of the trials were rated good quality and four were rated fair (Evidence Table 6).

The clinical outcomes evaluated were total mortality, CHD mortality, nonfatal MI, CHD events and revascularization (Evidence Tables 7 and 8). Most trials were designed to address clinical outcomes, but two were designed to evaluate change in intimal medial thickness of the carotid artery^{16, 25, 26} and included clinical events only as secondary outcomes.

For studies with mixed populations (e.g. some participants had CHD and some did not), the trial was classified as primary or secondary prevention based on the status of the majority of participants. Participants in most of the trials classified as primary prevention were at high risk for CHD outcomes due to presence of CHD risk factors.

Three trials included participants with and without CHD. In the colestipol trial, only 20 percent of participants had CHD and this trial was classified as a primary prevention study.³¹ In the Heart Protection Study, 65 percent of participants had known CHD and remaining 35 percent had peripheral vascular disease, cerebrovascular disease or diabetes.¹¹ Because the majority of participants had CHD and those without CHD were also at very high risk for CHD events, this trial was classified as a secondary prevention study. A recently published trial included older participants with CHD and those at high risk for CHD in approximately equal numbers⁴⁰ and did not present results stratified by history of CHD. Because of the equal distribution, we are unable to classify this study as either primary or secondary prevention. In addition, we could not include the results of this trial for any specific coronary disease endpoint because the results for women were only given for the composite outcome of cardiovascular events (CHD mortality, nonfatal MI, fatal stroke and nonfatal stroke).

Findings

For each outcome, we assessed the effects of lipid lowering separately for primary and secondary prevention studies. We also calculated summary estimates based on the findings of all eligible studies, those that used a statin as the lipid lowering agent and those that were rated good quality (Evidence Table 8).

Seven trials assessed the effects of lipid lowering among women with CHD (secondary prevention)^{7, 8, 11, 16, 27, 29, 30, 32- 37} and included a total of 8,244 women. Two of these trials used clofibrate as the intervention,^{29, 30} while five used a statin.^{7, 8, 11, 16, 27, 32- 37} Both of the trials of clofibrate were rated fair,^{29, 30} while all of the statin trials were rated good quality. While seven trials provided data, three were small (22 to 124 women),^{16, 29, 30} two were mid-sized (576^{8, 34, 35} and 827 women^{32, 33}) and only two included more than 1,000 women (1,516^{27, 36, 37} and 5,082¹¹). Evidence was also limited because several of the trials reported results among women for only one or two of the five outcomes of interest (total mortality, CHD mortality, nonfatal MI, CHD events and revascularization).

Four trials assessed the effects of lipid lowering among women without prior CHD (primary prevention)^{9, 25, 26, 31, 38, 39} and included 7,673 women. One of these trials used colestipol as the intervention³¹ and the rest used a statin. Two trials^{31, 39} were rated fair and the other two good quality.^{9, 25, 26, 38} Three of these trials included about 1,000 women or less (441,^{25, 26} 997,^{9, 38} and 1,184³¹) and one included 5,051.³⁹ As with the secondary prevention trials, many of these trials reported results among women for only one or two of the five outcomes of interest.

Secondary Prevention

Total mortality. Two trials,^{7, 16, 32, 33} both using a statin as the intervention, reported the effect of lipid lowering on mortality among a total of 899 women with CHD (Evidence Table 7). One of these trials¹⁶ enrolled only 22 women, so that essentially all of the data regarding the effects of lipid lowering for secondary prevention of mortality in women comes from one trial that used a statin as the intervention.^{7, 32, 33} Neither of the two trials found a reduction in risk of mortality among women (Evidence Table 7), and the summary relative risk was 1.11 (95% CI 0.66-1.87) (Evidence Table 8).

CHD mortality. Five trials reported the effect of lipid lowering on CHD mortality among 1,646 women with CHD (Evidence Table 7). However, three of these trials were small,^{16, 29, 30} and two,^{7, 8, 32- 35} both using a statin as the intervention, provide most of the evidence regarding the effect of lipid lowering on CHD mortality in women with CHD. The findings of these two trials were consistent in showing a reduced risk of CHD death among women treated with lipid lowering compared to controls. The summary relative risk for secondary prevention of CHD mortality was 0.74 (95% CI 0.57-0.96), suggesting a 26 percent reduction in risk of CHD mortality (Evidence Table 8).

Nonfatal MI. Five trials reported the effect of lipid lowering on risk for nonfatal MI in 1,646 women with CHD (Evidence Table 7). Three of these trials were small^{16, 29, 30} and two trials, both using a statin as the intervention,^{7, 8, 32- 35} provide most of the evidence regarding the effect of lipid lowering for secondary prevention of nonfatal MI in women. Both of these two trials showed a reduced risk for nonfatal MI and the summary relative risk was 0.64 (95% CI 0.50-0.82), suggesting a 36 percent reduced risk (Evidence Table 8).

CHD events. Four trials, all using a statin as the intervention, reported the effect of lipid lowering on CHD events in 8,001 women with CHD (Evidence Table 7). These trials consistently found a reduced risk of CHD events among women,^{7, 8, 11, 27, 32- 37} with a summary relative risk of 0.79 (95% CI 0.72-0.88), suggesting a 21 percent reduced risk of CHD events among women with CHD (Evidence Table 8).

Revascularization. Two trials, both using a statin as the intervention, reported the effect of lipid lowering for secondary prevention of revascularization procedures in 1,403 women with CHD (Evidence Table 7). Both of these trials found a reduction in risk among treated women and the summary relative risk was 0.70 (95% CI 0.42-1.16). Although the summary relative risk suggests a 30 percent reduction in risk of revascularization, this finding was not statistically significant (Evidence Table 8).

Drug class and study quality. Only two studies, including a total of 221 women, addressed the impact of lipid lowering drugs other than statins.^{29, 30} Thus, evidence on the effect of non-statin drugs is limited. However, the summary ORs were similar for all outcomes when findings were restricted to those studies using a statin. Both of the studies that used a non-statin drug were rated fair quality,^{29, 30}and all five of the trials that used a statin were rated good quality. Thus, the summary ORs are also unchanged when the results are restricted to good quality studies.

Sensitivity analyses. One trial that we included with the secondary prevention studies enrolled a mixed population of persons with and without CHD and reported the effect of statin treatment on risk for CHD events.¹¹ Because 65 percent of the participants had prior CHD and the rest had vascular disease or diabetes, we included the results of this trial as secondary prevention. We also performed a sensitivity analysis excluding the results of this trial. The summary relative risk for secondary prevention of CHD events excluding the results of this trial was essentially unchanged (summary relative risk 0.74; 95% CI 0.61-0.91). We also performed a sensitivity analysis by adding the results of the cardiovascular disease outcomes from the PROSPER trial to the summary results for secondary prevention of CHD events.⁴⁰ The summary relative risk for secondary prevention of CHD events including the results of this trial was essentially unchanged (summary relative risk 0.76; 95% CI 0.72-0.87).

Primary Prevention

Total mortality. Four trials^{9, 25, 26, 31, 38, 39} reported the effect of lipid lowering on mortality among 7,673 women without prior CHD (Evidence Table 7). One of these trials³¹ used colestipol as the intervention, while the rest used a statin. One of the trials reported a lower risk of mortality in women treated with lipid lowering compared to controls, but the other three did not (Evidence Table 7). The summary relative risk for primary prevention of mortality was 0.95 (95% CI 0.62-1.46) (Evidence Table 8).

CHD mortality. Three trials^{9, 25, 26, 31, 38} reported the effect of lipid lowering on CHD mortality among 2,622 women without prior CHD (Evidence Table 7). One of these trials³¹ used colestipol as the intervention, while the other two used a statin. One of the three trials reported a lower risk of CHD mortality in women treated with lipid lowering compared to controls,^{25, 26} but the others did not. The summary relative risk for primary prevention of CHD mortality was 1.07 (95% CI 0.47-2.40).

Nonfatal MI. Two trials^{9, 25, 26, 38} reported the effect of lipid lowering on risk for nonfatal MI in 1,646 women without prior CHD (Evidence Table 7). Both of these trials used a statin as the intervention, and both found a reduced risk of nonfatal MI among women treated with lipid lowering. The summary relative risk for primary prevention of nonfatal MI was 0.61 (95% CI 0.22-1.68). Although the summary relative risk suggests a 39 percent reduction in risk of nonfatal MI among treated women, this finding was not statistically significant.

CHD events. Two trials, both using a statin as the intervention,^{9, 38, 39} reported the effect of lipid lowering on risk for CHD events in 6,048 women without prior CHD (Evidence Table 7). The results of these trials are inconsistent, and the summary relative risk for primary prevention of CHD events was 0.87 (95% CI 0.50-1.49).

Revascularization. Only one trial reported the effect of statin therapy for primary prevention of revascularization procedures in women^{9, 38} (Evidence Table 7). This trial found a relative risk of 0.87 (95% CI 0.33-2.31).

Drug class and study quality. Evidence on the primary prevention effects of drugs other than statins is limited as only one trial addressed the impact of a non-statin drug.³¹ The summary ORs were similar for all outcomes when findings were restricted to those studies using a statin or to studies rated good quality.

Sensitivity analyses. We performed a sensitivity analysis by adding the results of the cardiovascular disease outcomes from the PROSPER trial to the summary results for primary prevention of CHD events.⁴⁰ The summary relative risk for primary prevention of CHD events including the results of this trial was essentially unchanged (summary relative risk 0.97; 95% CI 0.84-1.12).

Assessments for Heterogeneity and Publication Bias

There was no statistical evidence of heterogeneity in any of the overall summary estimates of the effect of lipid lowering on any outcome except for secondary prevention of revascularization (Evidence Table 8). There was no evidence of publication bias in any of the summary estimates.

Data Sources

We searched PubMed^®, the Cochrane Database, and DARE for studies in English or other languages published from 1966 through January 2002 . We also reviewed bibliographies and asked peer reviewers (Appendix A) to identify additional articles. In the case of multiple publications from a single study, we used the most comprehensive or recent publication.

Search Terms

Search terms were developed in collaboration with a medical librarian and include the following:

Inclusion Criteria

To be included, articles were required to fit the following criteria:

• 1

  Include both men and women and provide an estimate of the CHD risk associated with diabetes in both sexes.

• 2

  Followup of the cohort for at least six months.

• 3

  Data on one of the following outcomes: total mortality, CHD mortality, cardiovascular disease (CVD) mortality or nonfatal myocardial infarction (MI).

• 4

  Inclusion of primarily type 2 diabetic participants (defined by self-report, use of diabetic medication, medical record diagnosis, positive oral glucose tolerance test or an elevated fasting glucose).

• 5

  Inclusion of multivariate adjustment for confounders, including at least age, hypertension, hypercholesterolemia and smoking.

• 6

  Inclusion of a nondiabetic, concurrent control group.

• 7

  Published between January 1, 1966 and January 1, 2002. Articles published outside this date range that were recommended by peer reviewers (Appendix A) were included.

Definition of Outcomes

All included studies defined CHD mortality by the International Classification of Diseases, Ninth Revision (ICD-9) codes of 410 through 414 or by physician documentation of sudden cardiac death. Nonfatal MI was defined by definite electrocardiographic criteria using the Minnesota code, enzyme levels consistent with MI, self-report (with or without Rose questionnaire criteria), or medical record documentation.

Article Identification

An initial search using the terms listed above identified articles that potentially provided evidence. Two University of California, San Francisco (UCSF)-Stanford Evidence-based Practice Center (EPC) investigators reviewed the titles and excluded those that clearly did not provide data on humans or clearly did not address the question.

The abstracts of the remaining articles were reviewed independently by two UCSF-Stanford EPC physician investigators and coded using the categories listed below. Disagreements were discussed and consensus codes were entered into a database (Access, Microsoft Corporation).

RQ Research question: the article clearly does not address the research question

R Review – the study is a review that does not contain primary data

NH No humans - the study clearly does not include data on humans

O Outcome- the study clearly does not address the outcomes of interest

P Predictor- the study clearly does not include type 2 diabetics

E₁ Eligible – the study may contain primary evidence regarding the research questions in women and will be reviewed in full-text

Articles coded E₁ were retrieved and the full text was reviewed independently by two UCSF-Stanford EPC physician investigators. Names of authors and titles of journals were obscured before articles were reviewed.

Obtaining Unpublished Results in Women

Some eligible studies included women in the study population, but did not report findings separately by gender. In these instances we twice attempted to contact the authors of these studies to obtain these data. We contacted the authors of 26 articles, requesting the required information.^{4, 5, 8, 11, 15- 36} Authors of seven studies^{15- 21} provided the data necessary to satisfy inclusion criteria. Some authors were unable to recreate their original analyses ^{4, 11, 22, 23} or did not have the necessary variables in the dataset,^{24- 26} and others did not provide the requested data.^{5, 8, 27- 36}

Quality Assessment

The full text of each eligible study was reviewed independently by two UCSF-Stanford physician investigators who completed a quality evaluation form (Appendix B). Most of the studies included in this systematic review are prospective cohort studies. The major quality issues with this study design are lack of information on potential confounders, inadequate duration of followup, non-blinded outcome adjudication and loss to followup.

To be categorized as good quality, articles were required to meet the following parameters:

• Prospective cohort design (vs. retrospective cohort or cross-sectional design)

• Type 2 diabetes defined by fasting plasma glucose or oral glucose-tolerance test (vs. other definitions of diabetes)

• Multivariate adjustment for potential confounders in addition to age, hypertension, hypercholesterolemia, and smoking

• At least 14 years of followup time (the median length of followup of all studies)

• Less than 10 percent loss to followup

Studies were considered to be of fair quality if they met the following parameters:

• Retrospective or cross-sectional study design

• Criteria other than fasting plasma glucose or oral glucose-tolerance test used to define diabetes

• Adjusted for age, hypertension, hypercholesterolemia, and smoking only

• Follow-up time of less than 14 years

• More than 10 percent loss to followup

Data Abstraction

Two UCSF-Stanford EPC physician investigators independently reviewed the full text of each eligible study and completed a data abstraction form (Appendix C). One author reviewed titles and abstracts of articles retrieved from the search and excluded case reports, letters, comments, reviews, and reports without primary data. Two UCSF-Stanford EPC physician investigators reviewed the 50 remaining manuscripts to determine study eligibility. Data were extracted on study quality, participant characteristics, length of followup, and outcomes (CHD mortality, nonfatal MI, and cardiovascular or all-cause mortality). Discrepancies between reviewers were resolved by consensus. For studies with multiple publications, only data from the most comprehensive or recent publication were used.

Data Management and Archive

We entered all identified titles and abstracts in an EndNote^® (Niles Software, Inc) file that includes searchable key words as codes for eligibility. Information on all articles that were reviewed in full text was transferred from EndNote^® (Niles Software, Inc) to a database (Access, Microsoft^® Corporation) that allows us to categorize each article by reason for exclusion. Quality assessment data for each eligible study were also entered in the database (Access, Microsoft^® Corporation), allowing us to categorize eligible articles by quality.

Abstracted data were entered into a database (EXCEL, Microsoft^® Corporation) for preparation of evidence tables and calculation of summary estimates, confidence intervals and tests of heterogeneity.

The full-text articles that were retrieved, and the abstraction forms for each article are filed by topic and question in Dr. Grady's offices at the UCSF Mt. Zion Women's Health Clinical Research Center.

Data Analysis

The primary outcome of each study was expressed as the most adjusted odds ratio (and 95 percent confidence interval) for CHD events among persons with diabetes compared to those without diabetes. Summary estimates of odds ratio and 95 percent confidence intervals were calculated using a general variance-based (confidence interval) method³⁷ that retains adjustment for confounding. We calculated summary odds ratios using both a fixed and random effects model.³⁸ Results were similar using both models and we report only summary odds ratios based on the random effects model. The significance level for all summary relative risks was set at 0.05. All estimates were assessed for heterogeneity using a Chi square test with the significance level set at 0.10.

Publication bias usually occurs if small studies with unremarkable findings (odds ratios for the association of diabetes and CHD risk around 1.0) are not published while small studies with markedly positive findings (high odds ratios) are published. We calculated the correlation between individual study weight (1/variance) and odds ratio using Kendall’s Tau (a nonparametric correlation coefficient) to assess potential publication bias.

Summary estimates for men and women were compared using the Z-test, with a two-tailed five percent level of significance. The main comparisons were repeated in subgroups defined by race/ethnicity (white, black, Latino, Japanese American, and Native American) and by study design (prospective cohort and cross-sectional analyses). Sensitivity analyses were performed to assess the effects of study quality and degree of adjustment for confounding on the outcome.

Results of Study Identification

Our searches identified 4,578 titles. Of the 233 articles that contained primary data, 50 were duplicative publications, 46 did not include a nondiabetic control group, 44 did not provide information about the outcomes of interest and 26 did not perform analyses based on diabetes status. Eight studies did not provide data stratified by sex and used ineligible study designs,^{39- 46} seven were hospital-based studies with followup less than six months,^{47- 53} nine included only patients with prior MI.^{54- 62} Seven studies were excluded because the study population consisted of a single sex only.^{63- 69}

Of the 36 remaining studies, ten met all inclusion criteria.^{12, 70- 78} Twenty-two did not publish adequately adjusted risk estimates,^{4, 5, 15- 31, 33- 35} two did not report 95 percent confidence intervals or p-values for adjusted results,^{2, 8} and two provided only combined outcomes of nonfatal and fatal CHD.^{11, 36} We contacted the authors of these 26 articles twice requesting the required information.^{4, 5, 8, 11, 15- 36} Authors of seven studies^{15- 21} provided the data necessary to satisfy inclusion criteria. Some authors were unable to recreate their original analyses ^{4, 11, 22, 23} or did not have the necessary variables in the dataset,^{24- 26} and others did not provide the requested data.^{5, 8, 27- 36}

Description of Eligible Studies

After receiving additional information from authors, 17 studies fulfilled all inclusion criteria (Evidence Table 9); 12 were prospective cohort studies^{12, 15, 20, 70- 78} and five were cross-sectional analyses.^{16- 19, 21} More than one publication provided data from the same cohort ^{70, 77} such that the meta-analysis includes adjusted findings from 14 distinct study populations.

Followup time in the 12 prospective cohort studies ranged from 5 to 32 years (mean approximately 14 years). Most of the studies enrolled middle-aged participants; one study enrolled only subjects older than 65.¹⁹ The 14 study populations included 6,235 diabetic participants (48 percent women) and 71,306 nondiabetic control subjects (52 percent women). In 7 of the 17 studies,^{12, 15, 16, 18, 19, 71, 79} all diabetics were type 2; the remainder of the studies included a few type 1 diabetics, but the majority were type 2.^{20, 21, 70, 72, 144, 73-78}

Summary of Results

Evidence Table 10 presents the multivariate-adjusted odds ratios (OR) by gender and ethnicity for CHD mortality, nonfatal MI, and all-cause mortality for each included study. Most of the studies show a higher OR for CHD mortality and for nonfatal MI due to diabetes among women compared to men. Outcomes for cardiovascular and all-cause mortality are mixed with approximately half of the studies showing a higher odds ratio for women than men.

Data Sources

We used the results of a previous search of troponin articles in unstable angina (through 1999)³ and supplemented this with a second search (through 2002) to identify gender specific rates of cardiac outcome (death or myocardial infarction) for patients with non-ST elevation acute coronary syndromes with and without elevated troponin levels. Because few published studies provided sex specific data, we also contacted a selected group of study authors directly. Peer reviewers (Appendix A) were asked to submit articles that provide evidence to address the questions.

Search Terms

We searched MEDLINE^® (1966-2002) and reviewed cited references of retrieved articles to identify relevant published studies. Our search criteria were (1) the text word troponin, and (2) the text words angina or unstable or myocardial infarction or ischemia. We also performed a search of the EMBASE database from 1990-1998, but did not find any additional articles fulfilling the study criteria. We contacted experts in the field of cardiac markers to identify large unpublished cohort studies.

Inclusion and Exclusion Criteria

To be included, articles were required to fit the following inclusion criteria:

• 1

  Clinical trial or cohort study

• 2

  Evaluate patients with suspected myocardial ischemia

• 3

  Evaluate the prognostic value of troponin levels in patients with non-ST elevation acute coronary syndromes

• 4

  Published between January 1, 1966 and January 30, 2002.

We excluded studies that only included patients with myocardial infarction. We also excluded case-control studies, articles that did not report mortality, and articles with followup limited to hospitalization.

Article Identification

Study selection was performed initially by title review (PAH). Candidate abstracts were then reviewed and selected for data retrieval.

Data Abstraction

Two independent reviewers abstracted data for each article on standardized electronic data forms. A third reviewer compared their results and settled any differences. At least one reviewer of the pair had clinical cardiology expertise and one had experience in critical appraisal. We recorded the outcomes of nonfatal myocardial infarction and death. These were combined to form the outcome of death or myocardial infarction. If outcomes at more than one time period were reported, we used the value closest to 30 days following presentation.

Obtaining Unpublished Results

A number of eligible studies included women in the study population, but did not report findings separately by sex. In these instances we attempted to contact authors of all large studies (defined as >300 patients or >10 deaths during followup) to obtain this data. We contacted ten authors regarding nine studies^{8- 17} and received data from five studies.^{8- 10, 12, 13, 16}

Quality Assessment

We performed double abstractions for each article. For data obtained directly from authors we asked for confirmation of the data we received. We determined if the following quality indicators were present for the studies: clear listing of exclusion criteria, statement of whether providers were or were not blinded to the troponin results (for clinical trials), clear definition of myocardial infarction, classification of death outcome as cardiac death or total death. If less than three of these indicators (or the two applicable to cohort studies) were present, a study was classified as poor; otherwise it was considered to be good quality.

Data Management and Archive

All abstracted and author provided data were entered and stored electronically (EXCEL, Microsoft^® Corporation). A citation of each article reviewed was archived using EndNote^® (Niles Software Inc.).

Data Analysis

We used standard random (DerSimonian-Laird) and fixed (Peto) effects methods to estimate summary odds ratios for the outcomes of death and myocardial infarction.^{18, 19} Because both fixed and random-effects summary estimates were similar, we report only the random-effects results. For studies with no events in a patient group, we added 0.5 to each cell of the study for the random-effects calculation. We tested homogeneity of study effect size using a standard Chi-square test with the Q statistic.¹⁹ Summary estimates for men and women were compared using the z statistic. Data are presented as summary odds ratios with 95% confidence intervals, with two-tailed P-values and statistical significance set at P < 0.05.

Results of study identification

A total of 1,049 articles were identified with the MEDLINE^® and EMBASE databases and citation reviews. We excluded 878 articles based on title or abstract because they did not evaluate the prognostic value of troponin in patients with non-ST elevation acute coronary syndromes. The remaining 171 articles were retrieved and reviewed, and 78 of these articles met all of the inclusion/exclusion criteria.

Eligible articles were then reviewed to determine whether they reported relevant data for women. Only three of the 78 articles reported sex and troponin specific outcomes of death or myocardial infarction following hospitalization. The three included studies reported data for 407 women and 774 men.

Since so few studies reported data for an analysis of prognostic value of troponin by sex, we contacted the authors of the nine largest of the 78 studies to request outcomes data partitioned by sex and troponin test results. We obtained unpublished gender specific data for 2,762 women and 3,296 men from the authors of five of the nine large studies. ^{8- 10, 12, 13, 16} Two investigators reported on different topics for the same population.^{10, 16}

Patient Characteristics

Patient characteristics are displayed by sex in Evidence Table 13. Women were consistently older than men. Women were less likely to be smokers and have a history of myocardial infarction, but were more likely to have hypertension and diabetes than men.

Study Report Characteristics

Evidence Table 14 shows characteristics of the studies and the source of data used in the analysis (i.e., abstraction of data from publication or supplied directly by a study author).

Most studies used the highest troponin value to determine if the threshold was reached. The thresholds used for troponin T ranged from 0.1 to 0.2 ng/ml. The majority of studies were clinical trials where the troponin evaluation was a sub-study.

Quality of Study Reports

All eight included studies were rated as “good” quality. One study did not clearly list exclusion criteria.²⁰ All trials noted that health care providers were blinded to the troponin results. All studies stated how myocardial infarction was defined and all reported whether deaths referred to total or cardiac deaths.

Troponin Values

The frequency of elevated troponin and outcomes for women and men for each study are listed in Evidence Tables 15 and 16. Among 3,169 women, troponin was elevated in 1,118 (35 percent). This value ranged from 18 percent in the 1998 study report by Antman⁹ to 49 percent for the study by Safstrom.⁶ Among 4,070 men, the troponin was elevated in 1,571 (39 percent). This value ranged from 18 percent in the study by Galvani²¹ to 64 percent for the study by Safstrom.⁶

Death

There were 103 deaths among 3,169 women (3.3 percent). Among 2,051 women with a negative troponin, 42 died (2.0 percent) compared to 61 (5.5 percent) who died following a positive troponin value (Figure 1). There were 129 deaths among 4,070 men (3.2 percent). Among 2,499 men with a negative troponin, 47 died (1.9 percent) compared to 82 (5.2 percent) who died following a positive troponin value (Figure 2). The summary odds ratio for death with an elevated troponin was 2.63 (95 % CI 1.75-3.95) for women and 2.83 (95% CI 1.92-4.17) for men (p=0.8 for difference in odds ratio between men and women).

Death or Myocardial Infarction

There were 256 deaths or nonfatal myocardial infarctions among 3,169 women (8.1 percent). Among women with a negative troponin, 117 died or had a myocardial infarction (5.7 percent) compared to 139 (12.4 percent) who died or had a myocardial infarction following a positive troponin value (Figure 3). There were 366 deaths or myocardial infarctions among 4,070 men (9.0 percent). Among men with a negative troponin, 180 died or had a myocardial infarction (7.2 percent) compared to 186 (11.8 percent) who died or had a myocardial infarction following a positive troponin value (Figure 4). The summary odds ratio for the combined endpoint of death or MI for patients with an elevated troponin was for 2.16 (95% CI 1.65-2.81) women and 1.50 (95% CI 1.20-1.88) for men (p=0.04 for difference in odds ratio between men and women).

Nonfatal Myocardial Infarction

There were 153 nonfatal myocardial infarctions among 3,169 women (4.8 percent). Among women with a negative troponin, 75 had a nonfatal myocardial infarction (3.7 percent) compared to 78 (7.0 percent) who had a nonfatal myocardial infarction following a positive troponin value (Figure 5). There were 237 nonfatal myocardial infarctions among 4,070 men (5.8 percent). Among men with a negative troponin, 133 had a nonfatal myocardial infarction (5.3 percent) compared to 104 (6.6 percent) who had a nonfatal myocardial infarction following a positive troponin value (Figure 6). The odds ratio for death with an elevated troponin was for 1.80 (95% CI 1.28-2.54) women and 1.06 (95% CI 0.8-1.41) for men (p=0.02 for difference in odds ratio between men and women).

Comparisons Between Men and Women

The summary odds ratios for death (Figure 7), death or myocardial infarction (Figure 8) and nonfatal myocardial infarction (Figure 9) were computed separately for women and men. There was no significant heterogeneity (p<0.05) for any of the six summary estimates (three per gender). These results show that women and men had a similar increase in risk of death associated with a positive troponin. However, the relationship between death or myocardial infarction and elevated troponin was stronger for women (p=0.05). This was due to a stronger risk of nonfatal myocardial infarction with a positive troponin for women than for men (p=0.02).