2012 Training Course in MCH Epidemiology
The MCHB at HRSA and CDC offered this Training Course in MCH Epidemiology in 2012 as part of their ongoing effort to enhance the analytic capacity of state and local health agencies.
This national program was aimed primarily at professionals in state and local health agencies who have significant responsibility for collecting, processing, analyzing, and reporting maternal and child health data. The course was geared to individuals with intermediate to advanced skills in using statistical and epidemiologic methods, preferably in MCH or a related field.
Index of Presentations:
- Problem-Oriented Needs Assessment
- Overview of Linear Models
- Linear Models II
- Linear Models III
- Multifactorial Population Attributable Fractions for Priority Setting
- Trend Analysis for Performance Measurement and Program Evaluation
- Introduction to Evaluation Methods
- Propensity Scores for Program Evaluation
- Fixed Versus Random Effects Models for Multilevel and Longitudinal Data Analysis
- Moving Beyond Odds Ratios: Estimating and Presenting Absolute Risk Differences and Risk Ratios
- Presenting Multivariable Results
- Recent Advances in Missing Data Analysis: Imputation and Weighting
- Problem-Oriented Needs Assessment
Length: 100 minutes
Presenter: William M. Sappenfield, MD, MPH
Annotation: In this lecture, Dr. Sappenfield covers the basic principles and steps to needs assessment for effective public health practice. He begins by explaining the planning and action cycle with an applied example of late prenatal care entry in South Carolina and draws attention to potential pitfalls in connecting assessment data to action. He then discusses the phases of needs assessment from problem identification and prioritization to deeper analysis of the problem and strategy selection. An exercise comparing criteria-based and Q-sort prioritization is included.
- Overview of Linear Models
Length: 110 minutes
Presenter: Deborah R. Rosenberg, PhD
Annotation: In this introduction to multivariable modeling, Dr. Rosenberg begins by covering the importance of multivariable statistics in helping to isolate unique impacts in a complex world. Independent effects and inference are often needed to quantify health determinants and have relevance to program planning and policy recommendations. She then reviews the preparatory steps that should be taken before building a multivariable model to assess confounding and effect modification in 2x2 tables and the conceptual prioritization of variables. The different types of models, including linear and generalized linear (logistic, binomial, Poisson), are covered with applied examples and SAS code/output for examining the association between smoking and birthweight.
- Linear Models II
Length: 127 minutes
Presenter: Deborah R. Rosenberg, PhD
Annotation: In this second lecture on multivariable modeling, Dr. Rosenberg covers the assessment of confounding and effect modification in adjusted models, the creation and use of “dummy” variables to assess the effects of categorical variables or interactions, how to test custom contrasts between various categories, and strategies for model building.
- Linear Models III
Length: 96 minutes
Presenter: Deborah R. Rosenberg, PhD
In the final lecture on multivariable modeling, Dr. Rosenberg covers logistic models with multi-category outcomes known as multinomial or polytomous logistic models. She explains the differences in interpretation and assumptions from models with ordinal (cumulative logit) and nominal (generalized logit) response variables and how to choose between these options. She closes by answering common questions regarding model selection for variables that could be modeled in multiple forms.
- Multifactorial Population Attributable Fractions for Priority Setting
Length: 96 minutes
Presenters: Deborah R. Rosenberg, PhD; and Kristin Rankin, PhD
Annotation: Population attributable fractions (PAFs) help to quantify the impact of a risk factor at the population level, being influenced by the magnitude of the risk factor’s effect and the number of people with the risk factor. Drs Rosenberg and Rankin explain in this lecture how to compute (PAFs) from adjusted measures of effects in multivariable models. Because a person can have multiple risk factors, the estimate of mutually exclusive PAFs is affected by the order in which a given risk factor is hypothetically removed. Dr. Rankin explains the methods that are used to average PAFs over the range of possible permutations in the order of risk factor removal. Variable selection, modeling strategies, and interpretation issues are also discussed through applied examples. The steps and inputs for multivariable PAF calculations are described in the accompanying excel spreadsheet.
- Trend Analysis for Performance Measurement and Program Evaluation
Length: 94 minutes
Presenter: Deborah R. Rosenberg, PhD
Annotation: Dr. Rosenberg first outlines the range of different goals from trend analysis including overall patterns, differences between time periods, populations, or geographic areas, and projections. She then covers various analytic issues and techniques for assessing statistical significance. SAS and Joinpoint are used to test changes in trends/slopes both over time and between groups with examples relevant for program evaluation, target setting, and general monitoring and surveillance.
- Introduction to Evaluation Methods
Length: 127 minutes
Presenter: Embry Howell, PhD
Annotation: Dr. Howell covers the purpose and principles of evaluation along with key steps and potential pitfalls in this introduction to evaluation methods. Program evaluations are necessary for accountability, quality improvement, and improved allocation of resources. Dr. Howell then outlines some key steps in conducting evaluations: stakeholder engagement, design, implementation, dissemination, and program change/improvement. Common terminology and designs for implementation and outcome evaluations, including qualitative and quantitative methods, are discussed with a focus on potential biases. A bibliography and guide to planning an evaluation are provided in the attached documents.
- Propensity Scores for Program Evaluation
Length: 95 minutes
Presenters: Deborah R. Rosenberg, PhD; and Kristin Rankin, PhD
Annotation: In this lecture, Drs. Rosenberg and Rankin cover an advanced technique involving propensity scores for use in program evaluation. The goal of propensity scores is to better control for confounding by observed characteristics related to program participation or other exposures that are not random. Dr. Rosenberg explains how propensity scores are calculated and can be used to better control for imbalance in covariates by exposure status than conventional regression models. The techniques for using the propensity scores (weighting, stratification, and matching) are covered along with tips for building the propensity score model. Dr. Rankin illustrates applied examples with SAS code in evaluating the impact of 1) medical home access on children’s health, and 2) adequate prenatal care on preterm birth. In the final section, propensity score limitations and challenges are discussed interactively with a list of references provided for further study.
- Fixed Versus Random Effects Models for Multilevel and Longitudinal Data Analysis
Length: 120 minutes
Presenter: Ashley H. Schempf, PhD
Annotation: In this lecture on techniques for multilevel and longitudinal analysis, Dr. Schempf first outlines the unique features of data that is clustered or nested over space or time including improved inference and complex error structures. She then explains the differences, benefits, and disadvantages of various analytic techniques including fixed and random effects models as well as generalized estimating equations (GEE) and hybrid approaches in terms of inference with an applied multilevel example examining racial disparities in birth outcomes. Several other multilevel (neighborhood and sibling) and longitudinal policy evaluation examples are presented with a summary of how to select among the various approaches.
- Moving Beyond Odds Ratios: Estimating and Presenting Absolute Risk Differences and Risk Ratios
Length: 89 minutes
Presenter: Ashley H. Schempf, PhD
Annotation: Dr. Schempf begins this lecture by presenting some of the limitations of odds ratios and ratio measures in general. Absolute measures of association (risk or rate differences) directly quantify numbers affected and can be more relevant for public health while ratio or relative measures can help to standardize comparisons across groups, time periods, and different indicators. She then covers estimation options for risk differences and ratios with both simple and complex survey data, providing code in both SAS and STATA. A discussion and critique of examples in the literature are included.
- Presenting Multivariable Results
Length: 88 minutes
Presenter: Kristin Rankin, PhD
Annotation: In this lecture, Dr. Rankin outlines helpful principles for effective presentation of multivariable results that can improve communication and ultimate translation of findings. She starts by covering some basic tenets with illustrations of both effective and ineffective presentations. Dr. Rankin then discusses the importance of disseminating results to varied audiences (researchers, program managers, policy makers) and how to distill information without removing important details that affect inference and interpretation (e.g. confidence intervals). In the final segment, there is a discussion and critique of examples in published literature (attached PDF).
- Recent Advances in Missing Data Analysis: Imputation and Weighting
Length: 114 minutes
Presenter: Elizabeth Stuart, Ph.D.
Annotation: In this webinar, Dr. Stuart outlines the types of missing data (item and unit non-response) and their patterns and potential biases (missing completely at random, missing at random, not missing at random). She then presents an overview of available techniques and assumptions for handling both item and unit non-response with an in-depth example of multiple imputation with chained equations for item missingness in a PRAMS dataset. Sample multiple imputation code for Stata, R, and SAS (IVEware) are provided in the attached pdf.
Updated: January 2014
