HADM 379 - INTRODUCTION TO EPIDEMIOLOGY COMPLETE STUDY GUIDE & EXAM NOTES

Unit 4: Biostatistical Concepts Basic epidemiology requires a basic knowledge of biostatistics which includes effective methods of displaying data, and understanding the calculations required to analyze data in order to establish relationships and validity of data. Data exists in two manners: 1. Numerical- discrete values such as counts or measurements • Can be binary or continuous 2. Categorical- resulting from classification based on qualities. Data can be displayed in many different fashions, each with their own benefits and weaknesses. Tables and Graphs • Effective way to display information • Titles should be descriptive and succinct • Graphs are clear and simple, memorable and can show complex relationships • Tables can display complex data with precision and flexibility, 20 do not require a lot of skill to construct, and can use less physical space for a given amount of information Pie Charts • Display how an entity is divided into its parts Spot and Rate Maps • Display geographical locations of cases or rates • Rate maps are shaded according to difference in values: prevalence, incidence or mortality are often shown Bar Charts • Best suited for displaying numbers that compare two or more categories of data (ie male to female) • Lengths of bars convey the essence of the comparison, so scale must be correct Line Graphs • Best suited for displaying the amount of change or difference in a continuous variable (y-axis), typically over time (x-axis) • Can be linear or logarithmic • Caution: assess these graphs for scale breaks, which can present misleading representation of data Frequency Distributions and Histograms • Display the proportion of observations falling within a certain interval • The shape of the curve itself is highly informative • Normal distribution: bell curve appearance, with 68% of values falling withing +/- 1 standard deviation, and 95% of values falling within +/-2 standard deviations of the mean, and identical curves on both sides. Summary numbers measure central tendency; what is the “average” of the sample, as well as variability: what is the variation between individuals in each sample. The measures of central tendency are 1. Mean- the sample average. Calculated by the sum of all values divided by the number of values 21 2. Median-the middle measurement after all values are put in order. This measurement is useful when there are outliers 3. Mode- the value which occurs most frequently The measures of variability show how individuals within a sample are different from each other and can be calculated based on the differences between members for all possible pairs in the sample or the difference between each observation in the sample with the sample mean 1. Variance- measure of dispersion among the scores in a given population 2. Standard deviation- characterizes dispersion among scores in a given sample or population 3. Standard error- if sample was repeated, would the mean remain the same Statistical inference is the process of taking a single random sample and using it to make judgements about the population as a whole, using the summary numbers. • Confidence intervals are a measure of certainty that an unknown value falls within a certain range of values. The calculations effectively represent the population • The p-value determines the probability that an observed difference may have occurred by random chance • Statistical power is the probability of rejecting the null hypothesis. That is, the probability of a true positive result, and not an event of random chance • T-tests determine whether a mean differsfrom the theoretical value or from another mean (in a different sampling) • ANOVA or analysis of variance determines if there us significant differences between variances of multiple samples • Chi-square or contingency tables are tests which determine whether an observed association between 2 categorical variables differs from the theoretical value or from each other • Correlation is a statistical measure of the association between two variables. Correlation can be positive or inverse. 22 • Regression estimates the influence of one or more independent variables on a dependent variable, allowing for control of confounders o Linear regression- dependent variable is continuous and has normal distribution o Logistic regression- when variable is binary, that is present or not Recall that sample size is an important consideration of study design. Sample size calculations are based on • Prevalence • Acceptable error • Detectable difference Meta-analysis is a tool for increasing the statistical significance of data by aggregating data from multiple but comparable studies (increases sample size) Unit 5: Causal Associations The study of causation is a foundation of epidemiology • There is seldom a single cause for disease, contributions are multifactorial • Causal factors can be arranged into a hierarchy from the most proximal to distal socio-economic factors • There are multiple criteria for judging the evidence of causality Causal inferences are made based on associations between two variables, but there are different levels of association: Spurious association: observed associations may not be real Indirect association:statistical association between a characteristic and a disease due to a common factor (confounding) Direct association: factor causes effect