280 likes | 430 Views
Presenting Statistical Aspects of Your Research. Analysis of Factors Associated with Pre-term Births in North Carolina. 2012 NC Birth Data Factors Related to Preterm Births.
E N D
Presenting Statistical Aspects of Your Research Analysis of Factors Associated with Pre-term Births in North Carolina
2012 NC Birth DataFactors Related to Preterm Births Goal: Identify factors related to preterm birth (PTB) by using cross-sectional data reported on n = 118,391 birth certificates for live births in North Carolina in 2012. Of primary interest is the potential relationship between maternal smoking and prenatal care with PTB. Other demographic factors such as mother’s race, age, and education level as well physiological factors such as hypertension, previous birth history, and diabetes will also be considered.
2012 NC Birth DataFactors Considered • Mother’s Race – White, Black, Hispanic, American Indian, Other • Mother’s Education – mother’s education level (ordinal) • Mother’s Age – mother’s age (years) • Marital Status – mother’s marital status (1 = married, 2 = single) • No Care – mother received no pre-natal care (Y or N) • Cig During – mother smoked during pregnancy (Y or N) • GDIAB – gestational diabetes (Y or N) • GHYPE – gestational hypertension (Y or N) • PPB – previous pre-term birth (Y or N) • Over+ - mother is overweight or obese prior to pregnancy North Carolina Vital Statistics -- Births 2012 (1/23/2012)http://hdl.handle.net/1902.29/11614 UNF:5:uHpa3Rf5Sx9jFVCGIXkFQg== Odum Institute for Research in Social Science [Distributor] V1 [Version]
Demographics You can summarize mother demographics for the n = 118,391 live births in North Carolina in 2012. Here I used the Analyze > Tabulate command in JMP to create a table of summary statistics. For putting together a paper or presentation however, copying and pasting output from JMP is unsatisfactory. Creating a table in Word and/or PowerPoint would make for a much cleaner presentation!
Demographics – Preterm vs. Full-term As pre-term birth is the outcome of interest demographic comparisons of these two populations can be a nice addition. You can assess statistical significance by using appropriate bivariate tests (here all p-values < .0001).
County Level Maps You can use the maps you created to examine potential county differences on theses factors and the PTB rates. These maps could be used in your descriptive analysis or to support your findings & recommendations in the results/discussion sections. Discussions of counties that stand out as you did for homework might add to your conclusions. Some of you did a particularly fine job of this! Map of mean number of prenatal visits by county.
Crude Odds Ratios and Relative Risks Some papers will report both Crude OR’s and Adjusted OR’s. The adjusted OR’s come from the multiple logistic regression model that all of you are fitting as part of your analysis. The crude OR’s will come by considering each factor marginally (e.g. preterm vs. previous history of premature labor). I am not necessarily advocating this for your analysis, but it is something to consider. Also, as these data are NOT from a case-control study, you can look at relative risks (RR) instead of OR’s to quantify effects marginally. Other epidemiological measures can be examined as well. For example the attributable risk or risk difference, the population attributable risk (PAR), or population attributable risk fraction (PAF).
Example: Crude OR’s and RR’s Table # – Crude RR’s and OR’s for pre-term birth for factors considered. p < .0001 for all factors
Measures of Population Impact Population attributable risk (PAR) represents the absolute difference between risk (or rate) in the exposed population and the risk (or rate) in the unexposed group. If we have estimates of the rates among exposed (r1), unexposed (r0) as well as the proportion of the population that is exposed (p), the PAR is defined as: Population attributable risk fraction (PAF) is the measure of the proportion of all cases in the given population that may be accounted for by the exposure. It can also be caused the "etiological fraction". If r is the estimated rate of the outcome in the total population, then the PAF is defined as: If we have estimates of the relative risk or rate ratio (RR) and proportion of exposed in the population (p), the PAF can be found as follows: Measures of population impact are mostly used for planning public health measures. For example this can be to predict the impact of a change in the distribution of various risk factors on the frequency or incidence of disease in a given population.
PAR & PAF : Smoking and Preterm Birth (NC Births - 2012) From the full NC Births (2012) data we have the following estimates: Thus, which is reduction in the incidence of preterm births if the population were entirely non-smokers during their pregnancies. The PAF is given by: .0028/.1002 = .02794thus roughly 2.8% of the cases in the population can be accounted for by maternal smoking during pregnancy. This statement should NOT be interpreted as causational.
PAR & PAF : Prenatal Care and Preterm Birth (NC Births - 2012) From the full NC Births (2012) data we have the following estimates: 3059 3.19 Thus, which is reduction in the incidence of preterm births if the population were entirely non-smokers during their pregnancies. The PAF is given by: .00344/.1002 = .0344thus roughly 3.44% of the cases in the population can be accounted for by lack of prenatal care. Again this statement should NOT be interpreted as causational.
Multiple Logistic Regression When fitting a multiple logistic regression model to study potentially relevant factors simultaneously, all effects are adjusted for the other factors in your model. OR’s are again used to quantify the effects, but these will generally differ from the crude OR’s we considered previously. These adjusted OR’s can be put in a table with the crude OR’s shown previously or be placed in a separate table. The paper by Lewis, et al. I sent you does the former.
Multiple Logistic Regression Level 1 / Level 2 Odds Ratio Prob>Chisq Lower 95% Upper 95% Black Am.Ind 1.1718685 0.0490* 1.0006787 1.3801187 HispAm.Ind 0.6510726 <.0001* 0.5274443 0.8044058 Hisp Black 0.5555851 <.0001* 0.4787285 0.6416147 Other Am.Ind 0.689538 <.0001* 0.5846837 0.8173755 Other Black 0.588409 <.0001* 0.5481531 0.6313537 White Am.Ind. 0.8225018 0.0189* 0.7030908 0.967722 White Black 0.7018721 <.0001* 0.667477 0.7381008 WhiteHisp 1.2633027 0.0011* 1.0961237 1.4633925 White Other 1.1928303 <.0001* 1.1152479 1.2765049 Am.Ind. Black 0.8533381 0.0490* 0.7245754 0.9993218 Am.Ind. Hisp 1.535927 <.0001* 1.2431537 1.8959349 Black Hisp 1.7999044 <.0001* 1.5585679 2.0888666 Am.Ind. Other 1.4502464 <.0001* 1.223428 1.7103264 Black Other 1.6994981 <.0001* 1.5838983 1.824308 Am.In White 1.2158028 0.0189* 1.0333546 1.4222914 Black White 1.424761 <.0001* 1.3548284 1.4981789 Hisp White 0.7915759 0.0011* 0.6833437 0.9123058 Other White 0.8383422 <.0001* 0.7833891 0.8966617 Adjusted OR’s for Mother’s Race, adjusted for maternal smoking, marital status, mother’s age, gestational diabetes, gestational hypertension, and previous history of premature birth, and mother’s educational level.
Multiple Logistic Regression Level 1 / Level 2 Odds Ratio Prob>Chisq Lower 95% Upper 95% 1 2 1.1242089 0.0003* 1.054830 1.1980772 1 3 1.1146727 0.0008* 1.0463947 1.1872249 2 3 0.9915174 0.7638 0.9378521 1.0481238 1 4 1.2774486 <.0001* 1.1802327 1.3827778 2 4 1.1363089 0.0005* 1.0579047 1.2206704 3 4 1.1460302 <.0001* 1.0758257 1.221146 1 5 1.2524747 <.0001* 1.140667 1.3759116 2 5 1.1140943 0.0149* 1.0211982 1.2161394 3 5 1.1236255 0.0041* 1.0374927 1.2178402 1 = Less than HS 2 = HS Grad/GED 3 = Some College 4 = Bachelor’s Degree 5 = Master’s or Ph.D. Adjusted OR’s for Mother’s Education - adjusted for maternal smoking, marital status, mother’s age, gestational diabetes, gestational hypertension, and previous history of premature birth, and mother’s race.
Multiple Logistic Regression Level 1 / Level 2 Odds Ratio Prob>Chisq Lower 95% Upper 95% MARITAL STATUSSingleMarried 1.1684874 <.0001* 1.1116343 1.2282365PRENATAL CARE NoYes 3.9606617 <.0001* 3.5588051 4.4022196MATERNAL SMOKING Smoker Nonsmoker 1.1859801 <.0001* 1.1131197 1.2629405GESTATIONAL DIABETESYesNo 1.2558757 <.0001* 1.1585019 1.3598159GESTATIONAL HYPERTENSIONYesNo 3.1812302 <.0001* 2.9803234 3.3939524PREVIOUS HISTORY OF PREMATURE BIRTH YesNo 3.2768289 <.0001* 2.9771507 3.6026212OVERWEIGHT OR OBESE NoYes 1.0430068 0.0455* 1.00083961.0869491 MOTHER’S AGE Mother’s Age1.026572 (per 1 yr.) <.0001* 1.02263 1.030519 Adjusted OR’s are adjusted for the other factors in the table and are also adjusted for mothers education and race.
Summary of Logistic Regression As previous research has shown, factors such as mother’s race, education level, gestational conditions (e.g. diabetes and hypertension), and previous history of preterm birth are all associated with preterm birth in the directions we would expect. In addition, we see that lack of prenatal care and smoking during pregnancy are associated with an increased risk of preterm birth. We will examine these factors in more detail.
Discussion – No Prenatal Care Lack of prenatal care is associated with many of the factors examined in our analysis. We see that in general minority mothers have the highest percentages of mothers with no prenatal care, particularly Blacks and American Indians. Single mothers have the highest percentages with no prenatal care. Less educated women also have the highest percentages with no prenatal care. Those without private insurance have the highest rates of no prenatal care. Same is true for mothers who smoked during pregnancy and unfortunately those with a prior history of preterm birth. Finally we see that over 5% of the women with preterm birth had no prenatal care during the course of their pregnancy.
Discussion – Maternal Smoking Smoking during pregnancy is also associated with many of the factors examined in our analysis. We see that American Indians have the highest rates of maternal smoking. Less educated women and single women also have the highest percentages of maternal smoking. Those participating in the WIC program have higher rates of maternal smoking. Those without private insurance have the highest rates of maternal smoking, particularly those on Medicaid. Same is true unfortunately for those with a prior history of preterm birth. Finally we see that over 13% of the women with preterm birth smoked during pregnancy.
Discussion – NC Perinatal Association Perinatal Regions in North Carolina
Urban vs. Rural Counties Are there differences between the birth outcomes and demographics for urban counties vs. rural one?
Urban vs. Rural Counties Urban counties have worse birth outcomes and prenatal care in general than rural counties which may seem surprising given that we might expect less access to health care in rural counties. However, we do see that maternal smoking is more prevalent in rural areas.