No doubt you’ve seen something like this hierarchy of studies from Duke University’s Introduction to Evidence-Based Practice site. In this hierarchy, the studies are classified by the type of sample used.
Types of Samples
In a case study or case report, the sample is one or two individuals who experienced something unusual or noteworthy.
In a case-control study, the sample is a group of individuals with the characteristic of interest (usually diagnosed with something) called the cases and another set of individuals who do not have that characteristic called the controls. The controls are matched to the cases on as many demographic and health criteria as possible to allow researches to identify meaningful differences between the groups (the differences that lead to the diagnosis). Here is an example of a case-control study: Identifying risk factors for recurrent cesarean scar pregnancy: a case-control study. Continue reading
When we compare two populations and find a difference in a health outcome, we want to look more closely to determine if that outcome difference is due to a difference in health related characteristics or non-health related characteristics. When the difference is due to something other than health characteristics, the difference is considered a disparity.
Disparities are particularly disturbing because it means something about the health system or society is causing differences in health outcomes. Globally, disparities in maternal health have to do with the distribution of health care resources. Industrialized countries have higher rates of cesarean delivery and lower rates of maternal mortality than developing countries. Within countries, disparities may be related to access to care (who can and cannot gain entry to the health system), quality of care (being provided with appropriate and timely treatment), or acceptability of care (individuals are willing to use the treatment options available). Continue reading
Today was the opening of the AWHONN Annual Convention. This is my first time at AWHONN, so I am curious about how they do things. So far I have found the conversations easier to start than at the ICM Congress — but remember, these are all American Nurses so we have a commonality in training, role, and language that makes meeting new people easier.
I am glad I am at this conference. To be honest, I feel like sometimes nurses get left out of maternal and child health discussion, especially in public health. This is unfortunate. Nurses are the largest group of health care professionals in the USA, and the initiatives that are making a difference in birth are coming out of nursing. AWHONN’s The Full 40 Weeks is a great example.
I am still pouring through the information I gathered at ICM, and need to focus on learning what I can while I’m here. But in a few weeks, after the Statistics Summer Series, I’ll come back to these conferences and share some of the great information I’m learning. In the mean time, head over to the AWHONN website and order yourself some of these great buttons.
Remember from Monday that the population is the group of individuals you would like to learn about. As midwives, we are often interested in specific groups of women. These groups might be defined by age (e.g. reproductive aged women), by condition (e.g. women living with HIV), or any number of other ways we can divide women into groups.
When you divide your population into two or more groups, you can look at the difference in incidence or prevalence of an event to find out if the groups have a different risk. When you look at the groups separately, it is called stratifying. Have a look at this article as an example: Maternal age and successful induction of labor in the United States, 2006-2010. Continue reading
ACNM sent out the call for Abstracts for the 60th Annual Meeting yesterday. If you are a member you should have received the email. Full details can be found at the ACNM site. If not, look for updates at the ACNM website.
Types of Observations
Data is simply a collection of variables, grouped by observation. In health care, the observation is usually a “case” or a “person”. Researchers make each observation in a variety of ways. They may use a survey, asking individuals to report specific things. They may use biomarkers, taking blood or saliva samples to look for interesting components. They may use administrative data, looking at all hospital discharges or insurance charges for an organization. They may use surveillance data, examining collections of government mandated reporting data such as birth certificates. Continue reading
I ran into an infographic titled A Breakdown of Birth in the U.S.A. I wanted to like it, but the information on the poster is misleading. It does make a great little demonstration of how to learn to question what you read.
If you haven’t opened the infographic yet, put it in a new window and lets talk about the problems.
Always ask yourself first where the data is coming from. Notice the infographic provides only three sources for the data (which are nearly impossible to read), but I find that highly suspect given the variety of information the infographic shares. This is troublesome because grabbing data to verify the numbers will be difficult. When I follow the links I find my suspicions correct, the data is from many sources.. So we are left without knowing what year the census data is from and wondering why only the birth defects data was taken from the CDC. This author didn’t actually check the third source to see if the data were presented properly from the articles used.
Costs of birth
The problem here is this presentation is a bit vague — and you are going to see vague presentations of data pretty often. Is this author trying to say the United States spends the most per birth, or the most overall? Or, we might wonder if the costs are out-of pocket expenses or sticker price or insurance reimbursement. Without being able to verify the source I would still guess this is true, but misleading. The United States has one of the highest spending on health care per person for just about EVERY health condition, why would childbirth be different?
The United States has the third largest population in the world, so we will be expected to have the third largest number of births per year, and at least the third largest overall cost of delivery care because of that. The countries with higher population are China and India. Only about half of the births in India are attended by a skilled birth attendant which makes considerable changes in the cost. China has a high rate of hospital delivery, but the health system is considerably different. For comparison, the World Bank says the US spends $8,895 on health care per capita while China spends $322. Big difference. So while this statement is probably not inaccurate, it isn’t helpful for understanding anything about birth in the United States.
It is interesting to me that the data is given for the costs of birth in a hospital because as a researcher I will tell you this is a very difficult number to get. Actually, any hospital cost data is very difficult to get. We know what Medicaid reimburses hospitals, but other insurance programs guard their cost data as proprietary. Cost data presented in research are estimates at best, and vary considerably based on the type and source of data. Also remember, the cost of the physician or midwife is often separate from the cost of the hospital making good data even harder to compile. Regardless, the data I have seen does not suggest a doubling in reimbursement or out of pocket costs for a cesarean birth, and I have never been able to find a valid source to confirm a doubled “sticker price” as the norm.
Infant Mortality really deserves a post of its own. While many birth advocates hold this statistic up as evidence we need to change our style of managing birth, data shows it is more related to the ignored problems of poverty and racism. Check out the data in this report to start getting an idea of what I am talking about.
The infographic then moves into a short social commentary, data about who should and should not be a mother. Except the data again is misleading. There is a huge difference between a 12 year old being pregnant and an 18 year old pregnant, but this statistic wraps them in together. You will see this happen pretty often as well, authors will use data groupings that provide shock value. When you look at the breakdown (See Table 2), you will see the number of births doubles with each increase in age so the total number of births to women less than 18 is about 87K, and more than half of those are to women 17 years old. The idea that an 18 or 19 year old woman should not be coupled and bearing children is a societal standard that doesn’t fit all cultures within the United States. And don’t make the mistake of reading unmarried as uncoupled.
But then the infographic moves into birth defects – and gives a statistic that is global rather than specific to the United States. The author changed scale without telling you. Be on the lookout for this type of problem in writing because it happens more often than you think. It makes the graphic confusing (how can there be 8 million babies with birth defects if there are only 4 million births?). A better statistic could have been found.
I’m not really sure what the rest of the information is supposed to tell us about birth, but it is added perhaps because the infographic creator wanted to fill more space.
Be a Better Reader
Those are my quick comments about this infographic. I hope that helps you see some of the issues involved in sharing data and helps you become a more critical reader of the statistics when they are presented.
Next time we will look deeper at how to understand the statistic by looking at the measure.
Most of the research we are interested in as midwives is inferential – meaning we draw conclusions about a group of people based on the results. However, descriptive statistics are still very helpful.
Descriptive statistics help us organize and summarize information. For example, the number of births attended by midwives is a descriptive statistic. We can break down the data by country or state/province and see differences between groups. In experiments, the descriptive statistics help us ensure the two study groups are similar.
In healthcare, statistics isn’t useful without epidemiology. Epidemiology is the study of patterns of illness and conditions. We use epidemiology to determine the causes of conditions, the effects of exposures and treatments and the patterns of spread for health issues.
In epidemiology, there are two terms to be familiar with for descriptive statistics: incidence and prevalence. Let’s review these first with a non-health outcome – the number of midwives.
Incidence is a measure of new cases of something. For example, according to the American Midwifery Certification Board, the incidence of first time candidates for the midwifery certification exam has increased from 297 in 2005 to 542 in 2013. This is the number of new midwives each year.
Prevalence is a measure of the total number of cases of something, this means the total pre-existing and all new cases. For example, according to the North American Registry of Midwives , the prevalence of certified professional midwives had increased from 624 in 2000 to 1828 in 2010. This is the total number of midwives, both new and existing.
Take a look at this article: Prevalence of Hepatitis B Virus Seromarkers in Young Adults Vaccinated at Birth; Impact on the Epidemiology of Hepatitis B Infection in Iran
In this study, the researches wanted to see if the infant immunization schedule for Hep B was successful at reducing infection rates. To do this, the researchers tested a group of young adults to see what types of Hep B antibodies they had (from vaccination, from cleared infection or from chronic infection). This means the researchers were looking at prevalence, or the total number of people who test positive for each particular type of antibody.
Birth Worker Survey
The Birth Worker Survey allows us to gather some descriptive statistics about the services offered by the readers of the Birthing Naturally website. Remember, our total was 31 responses. Here are descriptive statistics for the most commonly reported services.
Provide Midwifery Services: n= 6 (19%)
Provide Doula Services: n=24 (77%)
Provide Childbirth Education Services: n=16 (52%)
Provide Breastfeeding Education: n=15 (48%)
Provide Herb or Essential Oil Blending: n=5 (16%)
Provide Labor Photography: n=5 (16%)
Provide Placental Preservation: n=4 (13%)
In the next post, we’ll talk about some common problems with presenting data.
A big part of understanding the statistics in research is understanding what is behind the data. What the data actually is, and how you obtain the data determine the types of statistical tests you will use. If you have been reading research, you probably already have an idea about most of the terms we will talk about today. But just in case, today we will make sure everyone is on the same page.
The most basic element of data is the variable. A variable is a measurement that represents a characteristic. Some examples of variables would be age, gender, height, parity, estimated fetal weight, and fetal heart rate. Each of these characteristics will be different for different individuals. It is these differences that interest us. How are they different? Why are they different? Do the differences matter?
Not all variables can be treated the same statistically. Characteristics of the variable itself determine what types of statistics can be used. All variables will be either continuous or discrete.
A continuous variable can take any value between its minimum and maximum. Characteristics such as age, weight, and blood pressure are continuous variables. Continuous variables are measured with a number, and the differences between each number are proportional meaning the magnitude of difference between 1 and 2 is the same as the magnitude of difference between 3 and 4.
Some continuous variables are interval and others are ratio. The difference is that a ratio variable has a specific zero measurement which indicates there is none of that variable. A gravid 4 has had twice as many pregnancies as a gravid 2, and a gravid 0 has had zero pregnancies. This means gravidy is a ratio variable.
A discrete variable is any variable that is not continuous, meaning it is only able to take on specific values. Characteristics such as race, sex, and use of oxytocin in labor are discrete variables. There are several types of discrete variables that may be used in research. Discrete variables are also known as categorical variables.
Some discrete variables are dichotomous, meaning there are only two options. Usually the options are yes or no.
Some discrete variables are nominal, meaning the characteristic has multiple values but the values do not have a mathematical relationship. For example, state of residence is a categorical value. Each individual may have a different state of residence, but there is no inherent ranking of the states. Other categorical variables include race and ethnicity, marital status, and place of birth.
Other discrete variables are ordinal, meaning the characteristic has multiple values that have a mathematical order without a defined magnitude of difference. For example, the pain scale asks women to rate the pain they feel as a number between 0 and 10. In this scale an answer of 10 is more pain than an answer of 9, but the difference between 9 and 10 may not be the same as the difference between 2 and 3. Similarly, the measurement of a “9” may not be the same from woman to woman.
Birth Worker Survey
The Birth Worker Survey included all types of variables. Age is a continuous ratio variable. Gender is dichotomous. Race is nominal. The questions about beliefs about childbirth are ordinal. This matters, because you report information about the variables differently.
50% of the respondents indicated they provide childbirth education services (a dichotomous variable). Childbirth educators were asked about the hours per session taught, number sessions in a typical course, and the number of courses taught per year. These are all continuous variables. Researchers using continuous variables will report the maximum value, minimum value and the mean value.
Hours per Session: Minimum 1.5; Maximum 10; Mean 3.1
Sessions per Course: Minimum 1; Maximum 12; Mean 6.4
Courses per Year: Minimum 2; Maximum 20; Mean 6.4
Childbirth Educators were also asked about the organizations they trained or certified with. This data is discrete, so it should be reported as a frequency. Frequency is reported as both the number of responses, percentage of the total. In research, these values would also be listed with measures of spread (confidence interval), but we will talk about those in a future post. Here is a partial list of the most frequently cited organizations.
American Academy of Husband Coached Childbirth n=4 (16%)
International Childbirth Education Association n=3 (12%)
Lamaze International n=3 (12%)
Childbirth and Postpartum Professionals Association n=2 (8%)
Childbirth International n=2 (8%)
Hypnobirthing n=2 (8%)
Spinning Babies n=2 (8%)
In the next post we will learn about descriptive statistics.
When I first began to read original research papers, I would skim over the statistical part to get to the conclusions. I understood statistics enough to tell if group A and B were different, and since the rest didn’t make sense to me I skipped it. I’m a little wiser now, and have a strong base of statistical knowledge to inform my reading. Honestly, it changes the research reading completely.
Statistics is simply the study of numerical data – how to collect it, analyze it and interpret it. In research, we use statistical testing to determine if the results of a study show a true phenomenon or were the result of chance.
Statistics can be broken down into two broad categories – descriptive or inferential. Descriptive statistics allow us to organize and summarize information from data. Inferential statistics lets us use a sample to draw conclusions about a population.
The population is the group of every individual you are interested in. For example, you may want to know about all the women of childbearing age in the United States. Or, you may want to know about all the women of childbearing age in the United States who were born in Mexico and primarily speak Spanish in their home. Both are populations and would be legitimate for a study. As the researcher, you define what population the study will examine.
The sample is the group of individuals you are able to collect information about. In inferential statistics, this group of individuals allows you to make estimates about the population.
Answering a Question
We like statistics because it helps us to answer a question. But for statistics to be useful we need to create a very specific question. The question is about the relationship between two variables. The Independent variable is the characteristic of interest, often thought of as the exposure. The Dependent variable is the outcome that depends on the independent variable.
Birth Worker Survey
The Birth Worker Survey can give us an example of independent and dependent variables.
Starting with the basics, we had 31 completed surveys. This is helpful statistically because 30 is a target number for being able to make assumptions about the normality of the mean of a variable, but that gets very technical and beyond what you need to know. Just be aware that all my hounding you to respond gave us a unusable sample for analyzing.
Our n (or size of the sample) is 31. Of those 24 women (incidentally, the respondents were all female) reported they do now or did in the past work as a doula.
If you remember, the Beliefs about birth questions were ranked 1-5, with 1 being strongly agree and 5 being strongly disagree. The simplest method to compare the two groups is to take the mean of the scores (the average of the scored values). When we do this, we find the doula group has a mean score of 1.912, indicating the doula group agrees that women should have a doula. The non-doula group has a mean score of 2.286, closer to an indifferent score.
So, is the dependent variable related to the independent variable? The means were different, but we have such a small sample that using this test we don’t get a statistically significant difference. The p-value is only .365, and the confidence interval for the difference between the means goes from -1.189 to 0.4509. This data and this test have not given evidence of a difference in belief about women having a doula in labor between doulas and other birth workers.
If you got a little lost, don’t worry. This is just day one, we will take time to discuss all of this later.
Point to Remember
While statistics is helpful for identifying the difference between a true phenomenon and a random result, it is important to remember statistics are only one piece of the design of a study that help you determine if findings are valid. The math can be good, and the result can be poor simply because the wrong sample was used or the wrong data were collected. We won’t go into all the aspects of a good study this summer, but perhaps we can plan for a series on research in the future. Later this week, we will start exploring data.
In the next post we will begin to explore the variable.