T de student y ANOVA presencial

Introduction to Bioestadística Class

Overview of the Session

• The session marks the second class of the bioestadística module, welcoming both in-person and online participants.

• A recap of descriptive statistics was provided, focusing on four types of variables: nominal, ordinal, discrete, and continuous. Methods for statistical and graphical representation were discussed.

Transition to Inferential Statistics

• The class shifts focus to inferential statistics or analytical statistics, which involves comparing two or more groups and associating independent and dependent variables.

• The instructor emphasizes the importance of understanding these concepts as foundational for future discussions.

Case Study Discussion

Introduction to Dr. Manetti's Clinical Trial

• Dr. Mauricio Manetti is introduced as an orthopedic surgeon conducting a clinical trial on a new opioid agonist (Ox 1004) versus morphine for post-operative pain relief after arthroscopy.

• The study design is described as multicenter, double-blind, randomized, and placebo-controlled with a primary outcome measure being pain intensity on a visual analog scale from 0 to 10 (continuous scale).

Sample Size Calculation

• A sample size of 100 patients (25 per center) was calculated to allow parametric tests even without normal data distribution assumptions.

• Dr. Roberto proposed using a non-paired test due to its simplicity for continuous outcome variables; however, this suggestion faced opposition regarding time considerations in post-operative analysis.

Statistical Concepts Introduced

Key Statistical Terms

• Important terms such as hypothesis testing (pruebas de hipótesis), ANOVA (análisis de varianza), linear regression (regresión lineal), paired vs unpaired data are highlighted during discussions among students about the case study findings.

Elements of PICO Question Framework

• Students discuss elements related to patient population (post-operative arthrosis patients), intervention (new drug Ox 1004), comparison group (morphine), and outcomes measured by pain scales. This framework aids in structuring research questions effectively.

Measurement Challenges in Pain Assessment

Pain Measurement Techniques

• Various methods exist for measuring pain; however, all have limitations due to individual differences in pain tolerance levels among patients which complicates objective assessments like visual analog scales used here.

Importance of Statistical Analysis

• Emphasis is placed on how statistical tests will be utilized based on whether data is treated as numerical or categorical when analyzing outcomes like pain levels across different treatment groups.

Understanding Hypothesis Testing

Definition and Purpose

• Hypothesis testing aims at determining if observed differences between groups can be attributed to chance through p-values that indicate significance levels.

Types of Tests Explained

• T-tests compare means between two groups while ANOVA extends this comparison across three or more groups; both are essential tools in inferential statistics.

Paired vs Unpaired Data

Distinction Between Data Types

• Clarification is made regarding paired data where measurements come from the same subjects versus unpaired data where measurements are independent across different subjects.

Preference for Independent Groups

• Most clinical studies favor independent group designs over paired ones due to potential confounding factors affecting results if only one group is analyzed pre-and post-intervention.

Regression Models Overview

Bivariate vs Multivariate Analysis

• Bivariate analyses consider only two variables while multivariate models incorporate multiple predictors allowing researchers to account for confounding factors common in observational studies.

(t=872] Conclusion & Next Steps

Summary Remarks

• The session concludes with reminders about upcoming classes focusing on bivariate statistical tests before advancing into more complex regression models later in the course.

Understanding Parametric and Non-Parametric Tests

Introduction to Statistical Tests

• The discussion begins with the distinction between parametric and non-parametric tests, emphasizing that parametric tests are used under specific criteria.

• It is noted that data must be on interval or ratio scales for parametric tests, highlighting the importance of numerical variables in statistical analysis.

Criteria for Parametric Tests

• Key requirements for conducting a parametric test include having a normal distribution of data, as illustrated by a Gaussian curve from previous discussions.

• Samples must be randomly obtained from a defined population, and variances across samples should be equal (homoscedasticity), which is crucial for valid results.

Choosing the Right Statistical Test

• Factors influencing the choice of statistical tests include data classification (continuous vs. categorical), distribution type (normal or not), number of groups, and whether groups are independent or related.

• A comparison example is introduced involving two groups: oxy 004 versus morphine, leading to considerations about appropriate statistical tests like Student's T-test.

Types of Statistical Tests

Student's T-Test

• The T-test is identified as suitable for comparing means between two independent groups; it can be paired or unpaired based on group relationships.

Non-parametric Alternatives

• The non-parametric equivalent to the T-test is mentioned as the Mann-Whitney U test, which serves similar purposes without strict assumptions about data distribution.

Chi-Square Test

• For categorical variables with two independent groups, the Chi-square test is introduced as another option for analysis.

Comparing Multiple Groups

ANOVA Overview

• When comparing three or more groups, ANOVA (Analysis of Variance) becomes necessary instead of using multiple T-tests due to increased risk of Type I error.

Non-parametric ANOVA Alternative

• The Kruskal-Wallis test serves as a non-parametric alternative to ANOVA when assumptions about normality cannot be met.

Correlation and Regression Analysis

Correlation Types

• Two types of correlation are discussed: Pearson’s correlation (parametric) and Spearman’s rank correlation (non-parametric), both useful in assessing relationships between continuous variables.

Regression Analysis

• Regression analysis is introduced as a multivariate approach where one variable predicts another; linear regression applies when both dependent and independent variables are numeric while logistic regression applies when they are nominal.

Application Example: Student's T-Test

Research Question Formulation

• An example research question regarding drug effectiveness in reducing blood pressure illustrates how to define outcomes clearly within statistical frameworks.

Variable Classification

• Blood pressure is classified as a continuous variable while treatment type represents a nominal variable with binary categories—essential for selecting appropriate statistical methods.

Statistical Comparison Process

Intuitive Approach vs. Statistical Methodology

• A discussion contrasts intuitive averaging methods against formal statistical approaches like the T-test that incorporate variability measures such as standard deviation into comparisons.

Importance of Variability Measures

• Emphasizing variability helps understand differences beyond mere averages; it accounts for individual variations within sample populations affecting overall results significantly.

Interpreting Results from Statistical Software

Understanding Output Data

• Outputs from software provide essential statistics including means, standard errors, confidence intervals, and p-values critical in hypothesis testing.

Hypothesis Testing Framework

• The null hypothesis posits no difference between group means; interpreting p-values informs whether this hypothesis can be rejected based on observed data differences.

This structured summary captures key insights from the transcript while providing timestamps linked directly to relevant sections for easy reference.

Understanding Box Plots and Statistical Significance

Introduction to Box Plots

• The speaker introduces box plots, explaining their components such as the boxes representing groups and whiskers indicating confidence intervals.

• A comparison is made between box plots and forest plots, highlighting that forest plots are used for association measures and dispersion.

Importance of P-values

• The speaker emphasizes the continued relevance of P-values in statistical analysis despite discussions about changing practices among educators.

• It is recommended to present at least two results from each test: a measure of magnitude (e.g., difference between groups) and a significance measure (P-value or confidence interval).

Hypothesis Testing

• Discussion on null hypothesis, where if zero is included in the confidence interval, it suggests that the null hypothesis cannot be rejected.

• The speaker checks for understanding among participants before moving on to further examples.

T-Test Fundamentals

Overview of T-Test

• Recap of T-tests as a method for comparing two means using measures of dispersion.

• An example study using NHFS data is introduced, focusing on various health-related variables.

Data Importing and Analysis Steps

• Instructions are provided on how to import data into Stata software for analysis.

• The first analytical steps include calculating average age and standard deviation from survey data.

Visualizing Data with Histograms

Creating Histograms

• Participants are guided through creating histograms to visualize age distribution among respondents.

• Discussion includes the principle of uncertainty and central limit theorem, allowing parametric tests even with non-normal distributions due to large sample sizes.

Comparing Groups Using T-Test

Statistical Comparisons Between Groups

• A question arises regarding significant differences in age between those who died versus those who survived; a T-test is deemed appropriate due to independent groups.

Conducting the Test

• Steps outlined for conducting a T-test include specifying numerical variables (age), categorical variables (mortality), and ensuring independence between groups.

Interpreting Results from T-Test

Analyzing Outcomes

• Results show an average age difference of 20.4 years between deceased individuals and survivors; this difference is statistically significant based on P-value findings.

Clinical vs. Statistical Significance

• Discussion highlights that while statistical significance may indicate differences in large datasets, clinical relevance must also be considered when interpreting results.

Exploring Research Questions

Formulating Research Questions

• A participant proposes research related to renal function improvement through dietary changes; discussion revolves around potential small but clinically relevant differences in outcomes.

ANOVA for Multiple Group Comparisons

Transitioning from T-Test to ANOVA

The session shifts focus towards comparing more than two means using ANOVA when dealing with multiple treatment groups affecting pain levels.

Setting Up ANOVA Tests

• Explanation provided about setting up hypotheses for ANOVA tests where null hypothesis states all group means are equal while alternative suggests at least one differs.

Problemas con el Error Tipo I en Comparaciones Múltiples

Introducción al Error Tipo I

• Hay una probabilidad de que los resultados obtenidos sean explicados por azar, lo que se traduce en un error tipo I. Esto es especialmente relevante cuando se realizan múltiples comparaciones.

• A medida que aumentan las comparaciones, la probabilidad de obtener al menos un resultado significativo por casualidad también aumenta, alcanzando hasta un 30% en algunos casos.

Corrección de Bonferroni

• Para mitigar el problema del error tipo I, es necesario realizar ajustes a los niveles de significancia mediante la corrección de Bonferroni, que es comúnmente aceptada y estricta.

• La corrección implica dividir el nivel de significancia (por ejemplo, 0.05) entre el número total de comparaciones realizadas para obtener un nuevo umbral más estricto.

Aplicación Práctica

• Al aplicar la corrección de Bonferroni, si se realizan seis comparaciones y se divide 0.05 entre 6, se obtiene un nuevo valor crítico (0.008). Solo los valores P menores a este serán considerados estadísticamente significativos.

• Se revisan varios valores P para determinar cuáles son estadísticamente significativos después del ajuste: solo uno resultó significativo tras aplicar la corrección.

Ejemplos y Discusiones sobre ANOVA

Importancia del ANOVA

• El ANOVA permite identificar si al menos una media es diferente entre grupos; sin embargo, no indica cuáles son esas diferencias específicas. Por ello, se requieren múltiples comparaciones posteriores para aclarar esto.

• Es fundamental ajustar por errores tipo I al realizar estas múltiples comparaciones debido a su incremento inherente con cada comparación adicional realizada.

Consideraciones sobre Datos

• Se discuten las condiciones necesarias para aplicar pruebas paramétricas como el ANOVA: muestras aleatorias seleccionadas e independencia entre observaciones son cruciales para asegurar validez en los resultados obtenidos.

Comparación de Grupos y Resultados Significativos

Análisis por Raza y Edad

• Se plantea una pregunta sobre si hay diferencias significativas en edad según raza utilizando datos categóricos (blanco, afroamericano y otras razas). La edad es considerada como variable numérica en este análisis.

• Las combinaciones posibles para comparar son limitadas a tres grupos independientes: blancos vs afroamericanos, blancos vs otras razas y afroamericanos vs otras razas; cada combinación debe ser evaluada individualmente para determinar diferencias estadísticas significativas.

Resultados Estadísticos

• Los resultados muestran diferencias promedio en edad entre grupos; sin embargo, no todas las diferencias resultaron ser estadísticamente significativas tras aplicar ajustes necesarios como la corrección de Bonferroni a los valores P originales obtenidos durante las pruebas iniciales del ANOVA.

Conclusiones Finales sobre Comparación Múltiple

Resumen del Proceso Analítico

• Se concluye que tanto el peso como otros factores pueden ser analizados usando métodos similares; sin embargo, siempre deben considerarse ajustes adecuados para evitar errores tipo I durante análisis complejos con múltiples variables categóricas o numéricas involucradas en estudios estadísticos avanzados como el ANOVA y sus extensiones posteriores con Bonferroni u otros métodos correctivos similares.