The tables that make up this section of Understanding Breast Cancer present the research that drives many of the recommendations and standards of practice related to breast cancer.
Research tables are useful for presenting data, but they can be hard to understand if you don’t work with them every day. They present a lot of information in a compact format.
Here, we present some basic concepts that may help you read and understand the research tables in this section.
The table below provides examples for this discussion. It has many of the features you will see in all of the tables.
The numbered table characteristics are described below.
Sample table – Table 3: Alcohol and breast cancer risk
Studies vary in how well they help answer scientific questions. When reviewing the research on a topic, it is important to recognize “good” studies. Good studies are well thought out (well-designed). Most scientific reviews set certain standards for the studies they include. These standards are called “selection criteria” and are listed for each table in this section.
The types of studies (for example, randomized controlled trial, prospective cohort, case-control) included in each table are listed in the selection criteria. Learn about the strengths and weaknesses of different types of research studies.
Selection criteria for most tables also include the minimum number of cases of breast cancer or participants for the studies in the table. When all else is equal, a larger number of people in a study means the study is better able to examine research questions. While there are large, poorly-designed studies, in general, large studies are better than small ones. Larger studies have more statistical power. This means the results from large studies are less likely to be due to chance than results from small studies.
Learn more about statistical power.
The first column (from the left) lists either the name of the study or the name of the first author of the published article.
A reference list is provided after each table so you can find the original published articles. Sometimes, a table will report the results of only one analysis. This can occur for two reasons. Either there is only one study that meets the selection criteria or there is one report that combines data from many studies into a single large analysis.
The second column describes the participants in each study. The study population of a randomized controlled trial is the total number of people who were randomized at the start of the study to either the treatment or the control group.
For prospective cohort studies, the study population is the number of people at the start of the study (baseline cohort). And, for case-control studies, it is the number of cases and the number of controls.
In some tables, more details of the study participants are included. For example, Table 11 on BRCA1 and BRCA2 gene mutations, has two columns describing the study populations. One column describes the characteristics of the people in the study and the other column shows the number of people in the study.
Length of follow-up
Randomized controlled trials and prospective cohort studies follow people forward in time to see who will have the outcome of interest (such as breast cancer or survival). For these studies, one column presents the length of follow-up time (that is, how many months or years people were followed in the study).
Because case-control studies do not follow people forward in time, there are data on length of follow-up for these studies.
Tables that focus on cumulative risk (discussed below) also can show the length of follow-up. These tables give the length of time, or age range, used to compute cumulative risk. Table 11, for example, shows the cumulative risk of cancer up to age 70 for people with a BRCA1 or BRCA2 gene mutation.
Some tables have columns with other information on the study population or the topic being studied. For example, Table 30b has a column with the age ranges for the study populations and Table 4 has a column with the comparisons of exercise used in the studies.
Sometimes, results for different groups or comparisons are presented. For example, the sample table has columns showing results for different comparisons of alcohol (for example, 1-2 drinks per day versus no alcohol and 2-4 drinks per day versus no alcohol). And, Table 4 has a column showing the results for premenopausal women and one for the results for postmenopausal women. This additional information gives more details about the studies and shows how the studies are similar to (and different from) each other.
Studies on the same topic can differ in important ways, such as length of follow-up and measures of a risk factor. Studies may look at outcomes among women of different ages or menopausal status. They may define “high” and “low” levels of a risk factor differently. These differences across studies are important to keep in mind when you review the findings in a table. They may help explain differences in findings.
Quantitative findings—understanding the numbers
All of the information presented in the tables is important, but the main purpose of the tables is to present quantitative findings—the numbers that show the risk linked to each topic.
These numbers are shown in the remaining columns of the tables. Before looking at these numbers, it is important to know what they represent. What is the outcome of interest—is it breast cancer? Is it five-year survival? Is it breast cancer recurrence?
Are groups being compared to each other? If so, what groups are being compared?
The headings of the columns with the quantitative findings help answer these questions.
Most often, quantitative findings are relative risks. For example, in Table 30b, the last column shows the relative risk of dying from breast cancer, comparing women who had screening mammography to those who did not.
A relative risk that is greater than one (1.25, for example) shows a factor increases risk. A relative risk that is less than one (0.75, for example) shows a factor decreases risk. And, a relative risk of one shows the factor neither increases nor decreases risk (this means the factor is not likely related to risk). Learn more about relative risk.
Most scientific studies report risk measures, such as relative risks, odds ratios and averages, with 95 percent confidence intervals (95% CI). A 95% CI around a risk measure means that there is a 95 percent chance that the "true" measure falls within the interval. Because there is random error in studies, and study populations are only samples of much larger populations, a single study does not give the “one” correct answer. There is always a range of likely answers. A single study gives a “best estimate” along with a 95 % CI of a likely range.
For relative risks and odds ratios, a 95% CI that includes the number 1.0 means there is no link between an exposure (like a risk factor or a treatment) and an outcome (like breast cancer or survival). When this happens, the results are said to be “not statistically significant.” This means that any link between the exposure and outcome is likely due to chance.
If a 95% CI does not include 1.0, the results are statistically significant and it is unlikely that the results are due to chance. Statistical significance is a key concept in health research and can be measured and presented in many ways. When findings are statistically significant, there is likely a true link between an exposure and an outcome.
A few examples from the sample table may better explain the concept of statistical significance.
The EPIC study found a relative risk of 1.07 with a 95% CI of 0.96 to 1.19. Women in this study who drank one to two drinks per day had a seven percent higher risk of breast cancer than women who did not drink alcohol. The 95% CI of 0.96 and 1.19 includes 1.0. This means these results are not statistically significant and the increased risk of breast cancer is likely due to chance.
The Million Women’s Study found a relative risk of breast cancer of 1.13 with a 95% CI of 1.10 to 1.16. Women in this study who drank one to two drinks per day had a 13 percent higher risk of breast cancer than women who did not drink alcohol. However, in this case, the 95% CI of 1.10 to 1.16 does not include 1.0. So, these results are statistically significant and suggest there is a likely true link between alcohol and breast cancer.
For any topic, it’s important to look at the findings as a whole. In the sample table above, most studies show a statistically significant increase in risk among women who drink alcohol compared to women who do not. Thus, the findings as a whole suggest that alcohol increases the risk of breast cancer.
The quantitative findings for some tables are presented as a cumulative risk (risk up to a certain age), often in the form of a percentage.
For example, Table 11 shows the cumulative risk of different types of cancer up to age 70 years for women who have a BRCA1 or BRCA2 mutations in their families. In the second study listed in this table (the Breast Cancer Linkage Consortium), 87 percent of women who came from families with a BRCA1 mutation had breast cancer by age 70 and 63 percent had ovarian cancer by age 70.
Sensitivity and specificity
Some tables show quantitative findings on the sensitivity of screening tests. The main goal of any cancer screening test is to correctly identify those people who have cancer (called the sensitivity of the test). When sensitivity is high, very few cases are missed. However, this means some healthy people will be misidentified as having cancer (a false positive result). For example, a sensitivity of 90 percent means that 90 percent of people tested who truly have cancer are correctly identified as having cancer.
An ideal cancer screening test would also be able to correctly identify all the people who do not have cancer as not having it (called the specificity of the test). When specificity is high, there are fewer false positive results, but more cases of true cancer are missed. For example, a specificity of 90 percent means that 90 percent of the people who are healthy are correctly identified as not having cancer.
No screening test has perfect sensitivity and perfect specificity. There is a trade-off between the two for all types of screening tests. That is, when a test gains sensitivity, it loses some specificity.
Summary relative risks from meta-analyses
Some tables have a summary relative risk reported at the bottom. A meta-analysis takes relative risks reported in different studies and “averages” them to come up with a single, summary measure. In Table 4, the summary relative risk comes from a meta-analysis.
Summary relative risks from pooled analyses
In a pooled analysis, the data from each person in each of the studies are combined into one large data set and analyses are done as if it were one big study. One of the summary relative risks at the bottom of Table 5 came from a pooled analysis.
A pooled analysis is almost always better than a meta-analysis. In a meta-analysis, researchers analyze results already published by others, which has limitations. In a pooled analysis, researchers combine the individual data from the different studies and this usually gives more statistical power than a meta-analyses. More statistical power means it is more likely the results are not simply due to chance.
In some cases, you may want to see more detail about a study than is given in the summary table. To help you find this information, the references for all the studies in a table are listed below the table. Each reference includes the authors of the study article, the title of the article, the year the article was published and the title and specific issue of the medical journal in which the article appeared.
PubMed, the National Library of Medicine's search engine, is a good source for finding summaries of science and medical journal articles (called abstracts). For some abstracts, PubMed also has links to the full text articles. Most medical journals have websites and offer their articles either for free or for a fee.
If you live near a university with a medical or public health school, you may be able to go to the school's medical library to get a copy of an article. Local public libraries may not carry medical journals, but they may be able to locate a copy of an article from another source.
More information on research studies
If you would like to read more about health research, a basic epidemiology textbook may be a good place to start. A medical librarian can help you find a textbook that fits your needs. Below are a few examples.
Aschengrau A and Seage GR, III. Essentials of Epidemiology in Public Health, 2nd edition (Jones & Bartlett Publishers, 2008)
Gordis L. Epidemiology, 4th edition (Elsevier Health Sciences, 2008)
Rothman KJ. Epidemiology: An Introduction (Oxford University Press, 2012)