Survival Data Analysis for Sekolah Tinggi Ilmu Statistik JakartaSetia Pramana
This document outlines a course on survival data analysis. It provides an overview of the course content which includes introduction to survival data, Kaplan-Meier survival curves, Cox proportional hazards models, parametric survival functions, and competing risks. It also describes the course workload which is 40% theory and 60% practice, including a group project, weekly presentations, and using R for analysis. Reference books and materials for learning R for survival analysis are also provided.
This document summarizes a presentation on statistical analysis of clinical cancer trial data. It discusses key topics like pivotal phase III cancer trial designs, endpoints, sample size considerations, statistical plans, interim analyses, and special issues. Superiority, non-inferiority, and equivalence designs are covered. Sample size calculations require understanding trial objectives, hypotheses, variability, and the minimum important difference to detect. Statistical plans address randomization, data collection and cleaning, and final analyses. Interim analyses balance early stopping for efficacy or futility while controlling type 1 errors. Lead statisticians and data safety monitoring boards oversee analyses.
Sample size and how to calculate it
- Why sample size is important
- Alpha and beta errors
- Main outcome and Effect size
- Practical examples using Means-Proportions-Correlation- Confidence Interval
1. The document summarizes key concepts in diagnostic test accuracy including sensitivity, specificity, predictive values, prevalence, and likelihood ratios.
2. It discusses ROC curves and how they are used to compare diagnostic tests by assessing the area under the curve.
3. Issues around bias in studies of diagnostic accuracy are covered such as spectrum, verification, and incorporation bias.
Evaluating a diagnostic test presentation www.eyenirvaan.com - part 2Eyenirvaan
This document discusses key concepts for evaluating diagnostic tests, including sensitivity, specificity, and predictive values. It explains how the criterion used to classify test results can affect sensitivity and specificity, as shown by receiver operating characteristic (ROC) curves. ROC curves summarize how criteria impact the trade-off between sensitivity and specificity. Likelihood ratios quantify this relationship and are used to calculate pre-test and post-test probabilities of disease. The sequential use of multiple diagnostic tests with different likelihood ratios can progressively update the probability that a patient has a disease.
This document provides an overview of statistical methods used in clinical research. It discusses different data types, descriptive statistics for summarizing data, standard error and confidence intervals. It also covers statistical tests such as t-tests, ANOVA, chi-squared tests, and non-parametric tests for comparing groups. Sample size calculations and the concept of type 1 and type 2 errors are also reviewed. The document serves as an introduction to common statistical analyses and concepts in clinical research.
Survival Data Analysis for Sekolah Tinggi Ilmu Statistik JakartaSetia Pramana
This document outlines a course on survival data analysis. It provides an overview of the course content which includes introduction to survival data, Kaplan-Meier survival curves, Cox proportional hazards models, parametric survival functions, and competing risks. It also describes the course workload which is 40% theory and 60% practice, including a group project, weekly presentations, and using R for analysis. Reference books and materials for learning R for survival analysis are also provided.
This document summarizes a presentation on statistical analysis of clinical cancer trial data. It discusses key topics like pivotal phase III cancer trial designs, endpoints, sample size considerations, statistical plans, interim analyses, and special issues. Superiority, non-inferiority, and equivalence designs are covered. Sample size calculations require understanding trial objectives, hypotheses, variability, and the minimum important difference to detect. Statistical plans address randomization, data collection and cleaning, and final analyses. Interim analyses balance early stopping for efficacy or futility while controlling type 1 errors. Lead statisticians and data safety monitoring boards oversee analyses.
Sample size and how to calculate it
- Why sample size is important
- Alpha and beta errors
- Main outcome and Effect size
- Practical examples using Means-Proportions-Correlation- Confidence Interval
1. The document summarizes key concepts in diagnostic test accuracy including sensitivity, specificity, predictive values, prevalence, and likelihood ratios.
2. It discusses ROC curves and how they are used to compare diagnostic tests by assessing the area under the curve.
3. Issues around bias in studies of diagnostic accuracy are covered such as spectrum, verification, and incorporation bias.
Evaluating a diagnostic test presentation www.eyenirvaan.com - part 2Eyenirvaan
This document discusses key concepts for evaluating diagnostic tests, including sensitivity, specificity, and predictive values. It explains how the criterion used to classify test results can affect sensitivity and specificity, as shown by receiver operating characteristic (ROC) curves. ROC curves summarize how criteria impact the trade-off between sensitivity and specificity. Likelihood ratios quantify this relationship and are used to calculate pre-test and post-test probabilities of disease. The sequential use of multiple diagnostic tests with different likelihood ratios can progressively update the probability that a patient has a disease.
This document provides an overview of statistical methods used in clinical research. It discusses different data types, descriptive statistics for summarizing data, standard error and confidence intervals. It also covers statistical tests such as t-tests, ANOVA, chi-squared tests, and non-parametric tests for comparing groups. Sample size calculations and the concept of type 1 and type 2 errors are also reviewed. The document serves as an introduction to common statistical analyses and concepts in clinical research.
This document summarizes complications of blood transfusion, including immediate and delayed immunologic and non-immunologic complications. Immunologic complications include acute hemolytic transfusion reactions, febrile nonhemolytic transfusion reactions, allergic/anaphylactic reactions, TRALI, delayed hemolytic transfusion reactions, and TA-GVHD. Non-immunologic complications include transfusion-transmitted infections, TACO, iron overload, and complications related to massive transfusion.
This document discusses common interview questions and answers related to clinical trial management jobs. It begins by defining key terms like clinical trials and their various types. It then addresses questions about participant eligibility, the trial process, informed consent, safety monitoring, and data management. Specific topics covered include trial phases, the purpose of placebos and control groups, adverse event reporting, and the responsibilities of clinical research coordinators.
The document discusses a study on sepsis in Indian patients admitted to the ICU. It finds that respiratory infections were the most common cause of sepsis. The study evaluated procalcitonin (PCT) levels to diagnose sepsis and found it to have 94% sensitivity. Higher PCT levels correlated with increased organ dysfunction as measured by SOFA scores. The study concludes PCT is a promising marker for diagnosing sepsis in critically ill patients that can help guide early management.
This document provides an overview of survival analysis. It defines survival analysis as statistical methods for analyzing longitudinal data on the occurrence of events over time. Key features include events that may or may not occur for subjects and the length of time until an event can vary. Censoring, where subjects drop out before an event, is accommodated. The objectives, terms, and reasons for using survival analysis are described. Key concepts like hazard rates, survival functions, and the Kaplan-Meier estimate are also introduced.
ANOVA is a statistical technique used to determine whether the means of groups are statistically different from each other. It can be used to establish cause-and-effect relationships with a certain degree of certainty. There are different types of ANOVA for different study designs. The basic parts of an ANOVA include sums of squares, degrees of freedom, mean squares, and the F-statistic. ANOVA can be performed in Excel using the data analysis tool. An example shows how ANOVA was used to analyze measurement data from multiple inspectors.
The document discusses various clinical trial designs, including parallel, crossover, dose-response, factorial, non-inferiority, and sequential parallel trials. It covers factors to consider when choosing a design like the questions being addressed and disease characteristics. Examples are provided for each design type to illustrate how they can be applied to evaluate different research questions. Issues related to active control and non-inferiority trials are also discussed.
- Probability theory describes the likelihood of chance outcomes and is measured on a scale from 0 to 1. Probability can be calculated classically based on equally likely outcomes or empirically based on relative frequency.
- Bayes' theorem allows updating probabilities based on new information by calculating conditional probabilities. It expresses the probability of an event A given evidence B in terms of prior probabilities and the likelihood of the evidence.
- The Monty Hall problem illustrates that switching doors in a game show scenario doubles the probability of winning the prize because it uses additional information provided by the host.
This chapter discusses chi-square tests and nonparametric tests. It covers chi-square tests for contingency tables to test differences between two or more proportions, including computing expected frequencies. The Marascuilo procedure is introduced for determining pairwise differences when proportions are found to be unequal. Chi-square tests of independence are discussed for contingency tables with more than two variables to test if the variables are independent. Nonparametric tests are also introduced. Examples are provided to demonstrate chi-square goodness of fit tests and tests of independence.
This document provides an introduction to various regression analysis techniques used in chemometrics, including partial least squares regression (PLSR), principal component regression (PCR), simple linear regression, and multiple linear regression. PLSR can be used to relate two data matrices and analyze data with many variables, while PCR reduces standard errors in regression estimates. Examples of applications in chemistry, medicine, food research, and pharmacology are given. Deming regression is described as a technique for fitting a line to data where both variables have measurement error.
ROC curves are used to evaluate machine learning algorithms and visualize the tradeoff between true positives and false positives. An ROC curve plots the true positive rate against the false positive rate for different discrimination thresholds. The area under the ROC curve (AUC) provides a single measure of performance, with higher values indicating better classification. While ROC curves are commonly used, precision-recall curves may provide a better evaluation for some applications by focusing on precision and recall rather than false positives.
The Chi Square Test is a widely used non-parametric test that does not rely on assumptions about population parameters. It compares observed frequencies to expected frequencies specified by the null hypothesis. The Chi Square value is calculated by summing the squared differences between observed and expected values divided by the expected values. The Chi Square value is then compared to a critical value based on the degrees of freedom. Common applications include tests of goodness of fit, independence of variables, and homogeneity of proportions.
演講-Meta analysis in medical research-張偉豪Beckett Hsieh
This document provides an overview of meta-analysis. It defines meta-analysis as a quantitative approach to systematically combining results from previous studies to arrive at conclusions about the body of research. It discusses key aspects of planning and conducting a meta-analysis such as defining the research question, searching for relevant literature, determining study eligibility, extracting data, analyzing effect sizes, assessing heterogeneity, and addressing publication bias. Software for performing meta-analyses and specific effect sizes like risk ratio and odds ratio are also mentioned.
The Seven Habits of Highly Effective StatisticiansStephen Senn
This document provides advice on habits that make statisticians effective. It discusses the importance of understanding causation, control, comparison and counterfactuals when thinking about effectiveness. It warns against proposing habits as causes without proper evaluation. Seven key habits are identified: read, listen, understand, think, do, calculate, and communicate. The document illustrates these habits through examples of invalid inversion, regression to the mean, and statistical mistakes. It emphasizes understanding concepts fundamentally rather than just mathematically and finding simple ways to communicate ideas.
The document discusses the history and development of blood transfusion, including major milestones like the establishment of the first blood bank in the US in 1937. It also covers risks associated with blood transfusion like infectious diseases and non-infectious complications. Guidelines are provided for rational use of blood and reducing exposure to allogeneic transfusion in surgical patients through various pre-operative, intra-operative and post-operative strategies. Thresholds for red blood cell transfusion are discussed based on recommendations from the American Association of Blood Banks. The importance of hospital transfusion committees and haemovigilance programs are also highlighted.
1) Statistics play an important role in medical research by describing diseases, making estimates from samples, determining significance of differences and associations, and making forecasts.
2) A statistician should be consulted at the planning, data collection, and reporting stages of research. At planning, they can help frame questions, determine sample size and sampling methods, and identify variables and scales of measurement.
3) It is important to utilize statisticians properly in research by involving them in the entire process and communicating effectively between clinical and statistical perspectives.
An overview of the ICH E9 guidance. Easy to follow, and I can provide a live presentation of this to your team! Great for those who are not familiar with statistics.
This document summarizes complications of blood transfusion, including immediate and delayed immunologic and non-immunologic complications. Immunologic complications include acute hemolytic transfusion reactions, febrile nonhemolytic transfusion reactions, allergic/anaphylactic reactions, TRALI, delayed hemolytic transfusion reactions, and TA-GVHD. Non-immunologic complications include transfusion-transmitted infections, TACO, iron overload, and complications related to massive transfusion.
This document discusses common interview questions and answers related to clinical trial management jobs. It begins by defining key terms like clinical trials and their various types. It then addresses questions about participant eligibility, the trial process, informed consent, safety monitoring, and data management. Specific topics covered include trial phases, the purpose of placebos and control groups, adverse event reporting, and the responsibilities of clinical research coordinators.
The document discusses a study on sepsis in Indian patients admitted to the ICU. It finds that respiratory infections were the most common cause of sepsis. The study evaluated procalcitonin (PCT) levels to diagnose sepsis and found it to have 94% sensitivity. Higher PCT levels correlated with increased organ dysfunction as measured by SOFA scores. The study concludes PCT is a promising marker for diagnosing sepsis in critically ill patients that can help guide early management.
This document provides an overview of survival analysis. It defines survival analysis as statistical methods for analyzing longitudinal data on the occurrence of events over time. Key features include events that may or may not occur for subjects and the length of time until an event can vary. Censoring, where subjects drop out before an event, is accommodated. The objectives, terms, and reasons for using survival analysis are described. Key concepts like hazard rates, survival functions, and the Kaplan-Meier estimate are also introduced.
ANOVA is a statistical technique used to determine whether the means of groups are statistically different from each other. It can be used to establish cause-and-effect relationships with a certain degree of certainty. There are different types of ANOVA for different study designs. The basic parts of an ANOVA include sums of squares, degrees of freedom, mean squares, and the F-statistic. ANOVA can be performed in Excel using the data analysis tool. An example shows how ANOVA was used to analyze measurement data from multiple inspectors.
The document discusses various clinical trial designs, including parallel, crossover, dose-response, factorial, non-inferiority, and sequential parallel trials. It covers factors to consider when choosing a design like the questions being addressed and disease characteristics. Examples are provided for each design type to illustrate how they can be applied to evaluate different research questions. Issues related to active control and non-inferiority trials are also discussed.
- Probability theory describes the likelihood of chance outcomes and is measured on a scale from 0 to 1. Probability can be calculated classically based on equally likely outcomes or empirically based on relative frequency.
- Bayes' theorem allows updating probabilities based on new information by calculating conditional probabilities. It expresses the probability of an event A given evidence B in terms of prior probabilities and the likelihood of the evidence.
- The Monty Hall problem illustrates that switching doors in a game show scenario doubles the probability of winning the prize because it uses additional information provided by the host.
This chapter discusses chi-square tests and nonparametric tests. It covers chi-square tests for contingency tables to test differences between two or more proportions, including computing expected frequencies. The Marascuilo procedure is introduced for determining pairwise differences when proportions are found to be unequal. Chi-square tests of independence are discussed for contingency tables with more than two variables to test if the variables are independent. Nonparametric tests are also introduced. Examples are provided to demonstrate chi-square goodness of fit tests and tests of independence.
This document provides an introduction to various regression analysis techniques used in chemometrics, including partial least squares regression (PLSR), principal component regression (PCR), simple linear regression, and multiple linear regression. PLSR can be used to relate two data matrices and analyze data with many variables, while PCR reduces standard errors in regression estimates. Examples of applications in chemistry, medicine, food research, and pharmacology are given. Deming regression is described as a technique for fitting a line to data where both variables have measurement error.
ROC curves are used to evaluate machine learning algorithms and visualize the tradeoff between true positives and false positives. An ROC curve plots the true positive rate against the false positive rate for different discrimination thresholds. The area under the ROC curve (AUC) provides a single measure of performance, with higher values indicating better classification. While ROC curves are commonly used, precision-recall curves may provide a better evaluation for some applications by focusing on precision and recall rather than false positives.
The Chi Square Test is a widely used non-parametric test that does not rely on assumptions about population parameters. It compares observed frequencies to expected frequencies specified by the null hypothesis. The Chi Square value is calculated by summing the squared differences between observed and expected values divided by the expected values. The Chi Square value is then compared to a critical value based on the degrees of freedom. Common applications include tests of goodness of fit, independence of variables, and homogeneity of proportions.
演講-Meta analysis in medical research-張偉豪Beckett Hsieh
This document provides an overview of meta-analysis. It defines meta-analysis as a quantitative approach to systematically combining results from previous studies to arrive at conclusions about the body of research. It discusses key aspects of planning and conducting a meta-analysis such as defining the research question, searching for relevant literature, determining study eligibility, extracting data, analyzing effect sizes, assessing heterogeneity, and addressing publication bias. Software for performing meta-analyses and specific effect sizes like risk ratio and odds ratio are also mentioned.
The Seven Habits of Highly Effective StatisticiansStephen Senn
This document provides advice on habits that make statisticians effective. It discusses the importance of understanding causation, control, comparison and counterfactuals when thinking about effectiveness. It warns against proposing habits as causes without proper evaluation. Seven key habits are identified: read, listen, understand, think, do, calculate, and communicate. The document illustrates these habits through examples of invalid inversion, regression to the mean, and statistical mistakes. It emphasizes understanding concepts fundamentally rather than just mathematically and finding simple ways to communicate ideas.
The document discusses the history and development of blood transfusion, including major milestones like the establishment of the first blood bank in the US in 1937. It also covers risks associated with blood transfusion like infectious diseases and non-infectious complications. Guidelines are provided for rational use of blood and reducing exposure to allogeneic transfusion in surgical patients through various pre-operative, intra-operative and post-operative strategies. Thresholds for red blood cell transfusion are discussed based on recommendations from the American Association of Blood Banks. The importance of hospital transfusion committees and haemovigilance programs are also highlighted.
1) Statistics play an important role in medical research by describing diseases, making estimates from samples, determining significance of differences and associations, and making forecasts.
2) A statistician should be consulted at the planning, data collection, and reporting stages of research. At planning, they can help frame questions, determine sample size and sampling methods, and identify variables and scales of measurement.
3) It is important to utilize statisticians properly in research by involving them in the entire process and communicating effectively between clinical and statistical perspectives.
An overview of the ICH E9 guidance. Easy to follow, and I can provide a live presentation of this to your team! Great for those who are not familiar with statistics.
Safety of Mebendazole Use During Lactationmothersafe
This case series study evaluated the safety of mebendazole use in 45 lactating women and their infants. Mebendazole was administered using single or repeated doses and was well tolerated by both the mothers and infants, with no adverse effects observed in infants. Mild GI irritability occurred in two treated mothers. This study provides the first evidence that mebendazole is safe for use in breastfeeding.
This document summarizes a study that examined the relationship between the drug domperidone, which is commonly used off-label to stimulate lactation, and the risk of ventricular arrhythmia and mortality during the postpartum period. The study used a retrospective cohort of over 225,000 women in British Columbia between 2002-2011. It found a possible doubling of the risk of hospitalization for ventricular arrhythmia among those exposed to domperidone, though the results were not statistically significant. Larger studies are needed to confirm any association.
This document summarizes key findings from several studies on exposures during pregnancy and lactation. The studies found:
- Teratogen information services receive thousands of calls annually regarding exposures to medications, infections, herbs and other substances during pregnancy and breastfeeding. The majority of calls concern analgesics, cold medications, herbs and dietary supplements.
- Most calls are made by exposed individuals themselves, highlighting a need for more education during prenatal care about risks of nonprescription drugs and vaccines.
- While some drugs used to treat autoimmune diseases like methotrexate and leflunomide can cause harm if taken during early pregnancy, others like sulfasalazine, azathioprine and antimalarials are
(마더리스크라운드) Thyroid disease in pregnancy 임신 중 갑상선mothersafe
This study examined pregnancy outcomes for women exposed to antithyroid medications or levothyroxine for thyroid disorders. The study found that infants of mothers treated for hyperthyroidism with antithyroid medications were more likely to be born preterm or with low birth weight. However, treatment of hypothyroidism with levothyroxine did not negatively impact birth outcomes and showed no increased risk of infant mortality. Additionally, the study found no evidence that levothyroxine exposure increased the risk of major congenital anomalies.
Diclectin in NVP, 44th 유럽기형학회보고 / 한정열 교수mothersafe
This document summarizes a presentation on making a difference as concerned scientists in an environmentally contaminated world. It discusses how scientists have identified problems like chemical pollution and investigated effects on human health. It provides examples like Rachel Carson's Silent Spring which brought attention to pesticide impacts and led to regulations. Endocrine disrupting chemicals can alter fetal development with impacts like reduced intelligence only appearing later. The document then summarizes several talks on topics like the impacts of maternal health conditions like diabetes and epilepsy during pregnancy, using the human placenta to test chemical safety, prenatal metal exposure and DNA methylation in the placenta, free fetal hemoglobin as a potential cause and target for preeclampsia, and revising guidelines
Maternal smoking during pregnancy was associated with an increased risk of major malformations in newborns. A study of over 2000 pregnancies found the risk of major malformations was 3.3 times higher for babies exposed to maternal smoking compared to non-exposed babies. Paternal smoking was also high among the smoking mothers, occurring in over 60% of smoking mothers compared to 38% of non-smoking mothers. Exposure to alcohol and lower education levels were also associated with increased risk of maternal smoking during pregnancy.
This document summarizes guidelines and studies on screening and management of subclinical hypothyroidism during pregnancy. Key points include:
- Guidelines from thyroid societies recommend trimester-specific reference ranges for TSH and treatment of SCH with levothyroxine.
- Studies show mixed results on associations between SCH and adverse pregnancy/child outcomes, and limited benefits of levothyroxine treatment.
- Targeted high-risk screening misses a significant percentage of women with thyroid dysfunction compared to universal screening.
- While evidence is still limited, most experts recommend universal screening to detect and treat overt hypothyroidism given potential benefits.
This document provides information about external cephalic version (ECV):
1. ECV is a procedure performed near term to manually turn a breech baby into a head-down position. The success rate of ECV is reported between 35-86%.
2. ECV has been performed since ancient times but was improved in the 1970s by performing it under tocolysis after screening with ultrasound and fetal monitoring. Recent studies show ECV effectively reduces non-cephalic births and C-sections for malpresentation.
3. Guidelines from obstetric organizations recommend offering ECV to women with a breech fetus near term due to evidence it can reduce C-sections. Factors like adequate amni
1. Proper management of diabetes before and during pregnancy is important to reduce risks of complications. Tight glucose control through medical nutrition therapy, exercise, and insulin treatment can decrease risks of fetal anomalies and growth issues.
2. Gestational diabetes is diagnosed through an oral glucose tolerance test and treated with lifestyle changes and possibly insulin to control blood glucose. Women with a history of GDM require follow up after pregnancy to screen for diabetes.
3. Preconception counseling and care is crucial for women with pre-existing diabetes to optimize health before pregnancy in order to lower risks during pregnancy through strict glucose monitoring and management.
This document discusses alcohol intake during pregnancy and fetal alcohol spectrum disorders (FASD). It provides statistics on alcohol use during pregnancy from various studies. It notes that a safe level of alcohol during pregnancy has not been determined, as the effects of alcohol on the fetus are variable depending on factors like the mother's metabolism and drinking patterns. Low to moderate prenatal alcohol exposure has not shown effects in some studies, but other research has found children with FAS even with reported low alcohol intake. The document describes clinical features of FASD including facial abnormalities, growth issues, central nervous system anomalies, and functional impairments. It discusses diagnostic criteria from various organizations and guidelines for diagnosing FASD.
1. 임상연구에 필요한 통계 분석 (2)
- 범주형 자료에 대한 분석 -
순천향대 중앙의료원 의학통계상담실
이 지 성
totoro96@schmc.ac.kr
Categorical data: 그 변수가 가질 수 있는 값이 명목형(nomial) 척도
또는 순위형(ordinal) 척도인 경우
명목형 척도: 혈액형(A, B, AB, O), 성별(남, 여)처럼 그 값들이 서로 다르
다는 것을 표현함.
순위형 척도: 상, 중, 하 또는 mild, moderate, severe 와 같은 증상이나 상
태의 심한 정도를 상대적으로 나타냄.
이러한 변수들이 갖는 각 값들을 범주(category)라고 함.
이들의 분포를 표현할 때에는, 각 범주에 속하는 상대적인 빈도(relative
frequency) 즉, 비율(proportion 혹은 rate)을 사용
분석방법: Chi-square test, Fisher’s exact test, McNemar’s Test, Kappa
statistic, Linear trend test, Cochran-Mantel-Haenzel Test 등
Introduction
2
2. 동질성 검정(Homogeneity test)
표본 수가 한 변수의 각 수준에 대해 미리 정해지는 경우
이 때 보고자 하는 것은 다른 변수에 대한 위 변수의 각 수준별 반응 분포
가 동일한가?
(예) 각 병원별 외과수술환자들의 사망률은 모두 동일한가?
두 범주형 변수들간 연관성 검정
3
병원
외과수술 결과
Total사망 생존
A 130 1970 2100
B 90 710 800
C 120 1380 1500
독립성 검정(Independence test)
전체 표본수가 정해지는 경우
이때 보고자 하는 것은 두 변수가 서로 관련이 없는가(즉, 서로 독립인가?)
(예) 교육수준과 소득수준은 서로 관련이 없는가, 즉, 서로 독립인가?
동질성 검정이든 독립성 검정이든 상관없이 모두 카이제곱 검정이라는 것을
사용함.
귀무가설: 두 범주형 변수간에 관련성이 없다(즉, 두 변수는 서로 독립이다).
두 범주형 변수들간 연관성 검정
4
교육수준
소득수준
상 중 하
대졸 255 105 81
고졸 110 92 66
중졸 90 113 88
3. A 22 Contingency table : a table composed of two rows cross-
classified by two columns
예:
예방접종여부와 인플루엔자 감염여부는 서로 독립적인가 아니면 서로 관련이 있는가?
만일 관련이 있다면, 예방접종을 받지 않으면 인플루엔자에 걸릴 위험이 증가하는가?
그렇다면 그 위험의 크기는 얼마나 되는가?
Pearson’s chi-square statistic
위의 검정통계량은 모든 칸에 대해서 기대빈도(mij)가 모두 5이상이어야 타당함.
각 칸의 관찰빈도(O)와 이에 해당하는 기대빈도 간의 차이가 크면 클수록 두 집
단의 비율은 다르다는 것을 의미.
1. Chi‐Square Test
5
예방접종
(exposure)
인플루엔자
TotalCase(=걸림) Control(안 걸림)
맞지 않음 n11 n12 n1+
맞음 n21 n22 n2+
Total n+1 n+2 n
n
nn
Ewhere
E
En
E
EO
ji
ij
i j ij
ijij
i j
2
1
2
1
2
1
22
1
2
1
2
2
~
6
H0: 예방접종 유무와 인플루엔자에 걸리게 될 사건은 독립이다.
H1: 두 사건은 서로 관련이 있다.
(귀무가설이 사실이라는 가정하에서) 기대빈도 계산
카이제곱 검정통계량
기각역: Reject H0 if χ2 > 3.84 (p-value=<.001)
결론: p-value = <.001 < = 0.05 reject H0 두 사건은 독립이 아니다.
예방접종
(exposure)
인플루엔자
TotalCase(=걸림) Control(안 걸림)
맞지 않음 80 140 220
맞음 20 220 240
Total 100 360 460
예방접종
(exposure)
인플루엔자
TotalCase(=걸림) Control(안 걸림)
맞지 않음 100×220/460=47.83 360×220/460=172.17 220
맞음 100×240/460=52.17 360×240/460=187.83 240
Total 100 360 460
2
1
2222
2
~01.53
83.187
)83.187220(
17.52
)17.5220(
17.172
)17.172140(
83.47
)83.4780(
5. 9
관찰빈도(observed frequency)=80
‐ 예방접종의 % = 36.4% = 80/220
‐ 인플루엔자의 % = 80.0% = 80/100
‐ 전체 % = 17.4% = 80/460를 각각 의미함
P‐value p‐값(유의확률) <.001은 유의
수준()으로 설정된 0.05보다
작으므로, 따라서 귀무가설
기각. 즉, 예방접종 여부와 인
플루엔자 여부는 관련이 있
다고 볼 수 있다.
† P-value by Chi-square test
인플루엔자 걸림 인플루엔자 안 걸림
예방접종
맞지않음
맞음
n (%) n (%) P‐value†
80
20
(36.4)
(8.3)
140
220
(63.6)
(91.7)
<.001
카이제곱검정의 타당성
2×2 분할표의 경우 (n은 Total number of observation)
n > 40 또는
20 < n < 40이면서, 각 칸의 기대빈도(expected frequency)가 모두 5 이
상일 때
r×c 분할표의 경우:
기대빈도가 5이하인 칸이 전체 칸의 20%이하이고, 1보다 작은 기대빈
도를 가지는 칸이 없을 때
자료가 위의 타당성 조건을 만족시키지 못할 때:
해당 행이나 열을 합하여 위의 조건들을 만족시키도록 함.
다음과 같은 경우에는 Fisher의 정확검정(Fisher’s exact test)을 사용함.
n < 20 또는
20 < n < 40이면서, 각 칸의 기대빈도 중 제일 작은 것이 5 이하일 때
카이제곱 검정의 타당성
10
6. Data concerning the possible association between high fat diet and the
risk of coronary heart disease
기대빈도 E11=13(8)/23=4.52, E21=10(8)/23=3.48
Two of the four cells have expected values less than 5. 즉, Data가 small 또는
zero cell count를 포함하고 있는 경우, 카이제곱 검정은 타당하지 않음.
Fisher의 정확검정(Fisher’s exact test)을 사용
2. Fisher’s Exact Test
11
Exposure
Heart Disease
TotalYes No
High Cholesterol Diet 11 4 15
Low Cholesterol Diet 2 6 8
Total 13 10 23
SPSS : Fisher’s exact test
12
자료: FatComp.sav
7. 13
Chi-square test
Fisher’s exact test
† P‐value by Fisher’s exact test
CHD=Yes CHD=No
Diet
High
Low
n (%) n (%) P‐value†
11
2
(73.3)
(25.0)
4
6
(26.7)
(75.0)
0.039
카이제곱검정 결과 유의한 차이가 있는 것으로 나타났으면 그 연관성
의 강도(the strength of an association)을 평가할 필요가 있음.
Cohort study(prospective)의 경우, RR을 incidence rate(발생율)에 대
한 relative risk measure로 사용
Case-control study(retrospective)의 경우,
Interest outcome이 Rare disease인 경우에는 OR을 relative risk measure
로 사용
Common disease인 경우에는 relative risk measure의 계산이 불가능함.
이 경우 OR은 단지 measure of association으로 사용할 수 있을 뿐임.
노출여부와 질병여부 간 연관성 측정
3. The strength of an association
14
Exposure
Disease
Total
Yes No
Yes a b a+b
no c d c+d
Total a+c b+d n
8. 15
Ex) 460명 대상, 예방접종 여부(아니오/예), 인플루엔자(걸림/안 걸림)
예방접종
인플루엔자
Total
걸림 안 걸림
맞지 않음 80 140 220
맞음 20 220 240
Total 100 360 460
예방접종여부와 인플루엔자 감염여부는 서
로 독립적인가 아니면 서로 관련이 있는가?
만일 관련이 있다면, 예방접종을 받지 않으
면 인플루엔자에 걸릴 위험이 증가하는가?
그렇다면, 그 위험의 크기는 얼마나 되는가?
카이제곱검정 결과 p-value<.001 reject H0
결과적으로 예방접종을 받지 않은 사람이 예방접종을 받은 사람에 비해 더 자주 인플
루엔자에 걸리게 된다고 말할 수 있다 (36.4% vs. 8.3%)
어느 정도나 더 자주 걸리게 되는가?(즉, 몇 배나 더 위험한가?)
Odds Ratio(OR) vs. Risk Ratio (RR)
16
- Case-control study: OR=(80/140)/(20/220) = 6.286
예방접종을 받지 않은 사람이 인플루엔자에 걸릴 odds는 예방접종을 받은 사람이 인플
루엔자에 걸릴 odds의 6.286배이다.
- Cohort study: RR=(80/220)/(20/240)=4.364
예방접종을 받지 않은 사람이 인플루엔자에 걸릴 risk는 예방접종을 받은 사람이 인플
루엔자에 걸릴 risk의 4.364배이다.
9. 전향적 연구의 경우
17
위험요인에 노출된 집단이 질병에 걸릴 Risk, R1=a/(a+b)
위험요인에 비노출된 집단이 질병에 걸릴 Risk, R2=c/(c+d)
비노출집단에 대한 노출집단이 질병에 걸릴 risk ratio, RR=R1/R2=a (c+d)/c(a+b)
노출집단이 질병에 걸릴 risk는 비노출집단이 질병에 걸릴 risk의 ‘RR’배이다.
Exposure
Disease
Total
Yes No
Yes a b a+b
no c d c+d
Total a+c b+d n
질병의 발생률(incidence rate) 파악이
가능한 연구설계
후향적 연구의 경우
18
질병에 걸린 집단이 위험에 노출될 odds, oddsD=a/c
정상인 집단이 위험에 노출될 odds, oddsND=b/d
정상집단에 대한 질병집단의 위험요인 노출 odds ratio, OR=oddsD/oddsND=ad/bc
질병집단이 위험요인에 노출된 odds는 정상집단이 위험요인에 노출된 odds의 ‘OR’배
이다.
노출중심으로 해석해보면,
• 위험요인에 노출된 집단이 질병에 걸릴 odds, oddsE=a/b
• 위험요인에 비노출된 집단이 질병에 걸릴 odds, oddsNE=c/d
• 비노출 집단에 대한 노출 집단의 질병에 걸릴 odds ratio, OR=oddsE/oddsNE=ad/bc
결국 OR은 같게 된다. 따라서 질병여부에 따라 설계된 연구이지만 해석은, odds의 개념을 이
용해서, “노출집단이 질병에 걸릴 odds는 비노출집단이 질병에 걸릴 odds의 ‘OR’배이다”라
고 해석
Exposure
Disease
Total
Yes No
Yes a b a+b
no c d c+d
Total a+c b+d n
10. 단면연구(cross‐sectional study)인 경우
19
단면연구에서는 노출여부와 질병여부간의 인과관계를 파악할 수 없음.
따라서 relative risk measure의 측정은 의미가 없다.
이 경우에는 질병의 발생률(incidence)가 아닌 유병률(prevalence)만이 파악될 수 있다.
즉, 유병률 비(prevalence ratio:PR)를 계산한다.
위험비(RR)을 계산한 뒤, 이 값을 PR 값으로 해석한다.
즉, ‘몇 배가 더 위험함’이 아닌 ‘몇 배 더 유병하고 있음’으로 해석한다.
Exposure
Disease
Total
Yes No
Yes a b a+b
no c d c+d
Total a+c b+d n
전체 인원 수 n명을 대상으로 노출여
부 및 질병여부를 파악한 연구설계
SPSS : 위험도분석
20
자료: 인플루엔자-coding1.sav
11. 앞 예제의 분석결과에 대한 해석
21
‘예방접종=맞음’인 경우에 비해, ‘예방접종=맞지 않음’인 경우에 인플루엔자가
걸릴 OR 및 해당 CI
‘예방접종=맞음’인 경우에 비해, ‘예방접종=맞지 않음’인 경우에 인플루엔자가
걸릴 RR 및 해당 CI
‘예방접종=맞음’인 경우에 비해, ‘예방접종=맞지 않음’인 경우에 인플루엔자가
걸리지 않을 RR 및 해당 CI
Matched case-control study
134 cases and 134 matched controls, for a total of 268 subjects.
Concordant pair(=13 pairs, 92 pairs)
No information about the association between risk factor and disease
McNemar’ test uses only the number of discordant pairs.
4. 맥니마 검정(McNemar’s Test)
22
Case
Control
Total+ -
+ 13 4 38
- 25 92 96
Total 17 117 134
2
1
22
2
~79.13
425
14251
SR
SR
결론: p-value = <.001 < =0.05 There is an association between risk factor and disease.
13. 일치성 척도 (Agreement measurement)
두 관찰자 간의 측정 범주값에 대한 일치성 정도를 측정하는 방법
두 관찰자의 평가가 우연히 일치할 가능성을 보정한 두 관찰자간의 일치도
5. Cohen’s Kappa
25
The Kappa Statistic의 이론적 배경
26
Cohen(1968)의 kappa 계수
Kappa, K = po – pe / 1 – pe
po : “observed” agreement proportion
= (15 + 70)/100 = 0.85
pe : by chance alone (“expected”
agreement proportion)
= [(n1/n)×(m1/n)]+[(n0/n)×(m0/n)]
= [(25/100)×(20/100)]+[(75/100)×(80/100)]
= 0.05 + 0.6 = 0.65
Kappa = (po ‐ pe)/(1 ‐ pe)
= (0.85 – 0.65)/(1 – 0.65) = 0.57
Moderate agreement
14. 두 산부인과 의사가 140명의 환자를 transvaginal ultrasonography를
시행하여 난소암의 여부와 정도를 진단하였다.
두 의사의 난소암 판정정도는 일치하는가?
Kappa 계수(Kappa coefficient)로 평가
Example
27
Doctor A
Doctor B negative + ++ 전이
negative 32 3 2 0
+ 5 10 11 0
++ 0 4 42 0
전이 0 1 15 15
SPSS : Kappa 분석
28
자료: Kappa.sav
분석을 시작하기 전에 count를
“가중설정”을 해 주어야 한다.
16. 예) Contingency Table Blood Pressure Data
In the form of 2×k contingency table, the rows have a distinct order(i.e.,
time points, ages, or doses), this information is not used in the standard
chi-square test.
The rows are ordered, and you may wish to ask whether there is a
linear trend.
That is, whether the prevalence of hypertension changes linearly with class.
Null hypothesis: There is no correlation between row (class) number
and the proportion of subjects who are hypertensive (in left row).
6. Chi‐Square Test for Trend
31
Class
Blood Pressure
High Not High
Ⅰ
Ⅱ
Ⅲ
Ⅳ
5
11
12
14
25
19
19
16
SPSS : Trend Test
32
분석을 시작하기 전에 count를“가중설
정”을 해 주어야 한다.
17. 33
Chi-Square Test for Trend(선형 대 선형결합): p-value = 0.018 There is a
significant linear trend among the ordered categories defining the rows
and the proportion of subjects in the left column.
Chi-square test for trend uses more information and tests a narrower set of
alternative hypotheses than does the chi-square test for independence.
Chi-Square Test : p-value = 0.091 The row and column variables are not
significantly associated.
SPSS output
34
18. 4개의 병원으로부터 호흡기 곤란 환자들을 기존 치료제와 새로운 치
료제에 랜덤하게 할당한 후 호전의 유무를 조사한 자료임. 치료제에
따라 호전도의 차이가 있는가이며, 병원의 효과를 제어하고 싶다
7. Cochran‐Mantel‐Haenzel Test
35
병원 치료제 호전 호전되지 않음
A
기존치료제
새로운 치료제
9
11
5
6
B
기존치료제
새로운 치료제
7
8
5
3
C
기존치료제
새로운 치료제
4
7
6
5
D
기존치료제
새로운 치료제
18
26
11
4
독립된 K개의 그룹이 있을 때, 그룹의 효과를 제어한 반응률의 차이가
있는지를 검정하는 방법
처리와 반응률의 자료가 독립된 여러 병원으로부터 얻어졌을 때, 병원을
하나의 층(strata)로 보고 이를 제어한 처리와 반응사이의 연관성을 알아
보는 경우
병원을 층(strata)으로 두고 두 처리간 반응률의 차이를 조사하는 것은 병
원에 따라 처리 간 반응률이 달라질 수 있기 때문임.
병원이라는 층변수(stratification variable)를 제어하면서 전체적인 반응
률의 차이를 조사하는 방법
따라서 K개의 층이 있고 처리 1에서의 반응율을 p1, 처리 2에서의 반응율
을 p2라고 두 처리 간에 전체적인 반응률의 차이가 있는지를 검정하고 함.
Cochran‐Mantel‐Haenzel (CMH) Test
36
20. 39
The significant p-value (=0.041) 병원을 통제한
결과, 치료제에 따라 호전여부는 다르다는 것은 통계
적으로 유의하다.
40
The Breslow-Day test : a method for the testing
for homogeneity of the odds ratio over multiple
strata Adjusted OR(=1.625) can be used.
The common OR(Mantel-Haenszel OR) = 2.147
95% CI: 1.019 – 4.520