Download presentation

Presentation is loading. Please wait.

Published byDeon Laker Modified over 6 years ago

1
How would you explain the smoking paradox

2
Smokers fair better after an infarction in hospital than non-smokers. This apparently disagrees with the view that smoking is bad for you health. (Appears not to make sense). Reasons: 1. Whether you smoke or not, does not effect the severity of the heart attack. The study does not look at heart attack prevention, it merely looks at what happens following a heart attack and the risk of death. In other words, it is the heart attack that causes the case fatality, not the smoking. 2. Higher percentages of smokers die before ever reaching hospital than non-smokers (smokers 38.4(odds ratio 1.09) compared with non-smokes 37.6 (odds ratio 1)). However the confidence interval does range from 0.93- 1.27, therefore significance not strong 3. The overall adjusted odds ratio on case fatality is 1 (non smokers), 0.97 (smokers) and 0.85 ( ex smokers), the confidence limits all crossing 1 implying the possibility of no significant difference. The Paradox:

3
Clear aims: important area of health education in reducing mortality from smoking. Clear aims: important area of health education in reducing mortality from smoking. Study type: case controlled study matching the risk of dying between smokers and non-smokers following an acute myocardial infarction. Study type: case controlled study matching the risk of dying between smokers and non-smokers following an acute myocardial infarction. Matching: study group of smokers is matched to a controlled group of non- smokers. Matching: study group of smokers is matched to a controlled group of non- smokers. Group selection: it is a register based study looking at all patients in Auckland with coronary related deaths. This reduces bias with regard to case selection. Group selection: it is a register based study looking at all patients in Auckland with coronary related deaths. This reduces bias with regard to case selection. Size: large number of cases (5106). Size: large number of cases (5106). Control matching: the mean age and age standard deviation (SD) is similar in all three groups. Control matching: the mean age and age standard deviation (SD) is similar in all three groups. Endpoint: death is a well-defined or specific endpoint Endpoint: death is a well-defined or specific endpoint Statistical analysis: a logistic regression model allows us to assess the importance of several variables simultaneously, in this case age, sex, previous heart disease, and previous MI, thereby reducing confounding Statistical analysis: a logistic regression model allows us to assess the importance of several variables simultaneously, in this case age, sex, previous heart disease, and previous MI, thereby reducing confounding Method: study criteria, methods of case finding and data collection procedures have been published. Method: study criteria, methods of case finding and data collection procedures have been published. Strengths

4
Case controlled study: lies lower down in the hierarchy of evidence Case controlled study: lies lower down in the hierarchy of evidence Case control study: shows association and not causation, Case control study: shows association and not causation, Relevance to me: do patients in Auckland New Zealand match my own patients Relevance to me: do patients in Auckland New Zealand match my own patients Socioeconomic data: no mention of racial characteristics and other characteristics. Socioeconomic data: no mention of racial characteristics and other characteristics. The Crude Odds ratios all cross 1, suggesting that no difference in fact exists. The Crude Odds ratios all cross 1, suggesting that no difference in fact exists. Other risk factors: no mention regarding weight, blood pressure, and cholesterol which could equally be factors relating to cardiac death. Other risk factors: no mention regarding weight, blood pressure, and cholesterol which could equally be factors relating to cardiac death. Control matching: the three groups are not evenly matched in all aspects. Control matching: the three groups are not evenly matched in all aspects. 10% difference in numbers of men between non smokers and ex-smokers 10% difference in numbers of men between non smokers and ex-smokers 25 % difference in absolute numbers between those with angina/no angina. 25 % difference in absolute numbers between those with angina/no angina. There are a large number of ex-smokes who have had MIs than in the group of non-smokers and current smokers. One powerful reason to stop smoking is having an MI, and therefore this group may be a biased group There are a large number of ex-smokes who have had MIs than in the group of non-smokers and current smokers. One powerful reason to stop smoking is having an MI, and therefore this group may be a biased group Weakness

5
Issues raised Case controlled trial Odds Ratios Confidence intervals, Significant tests (P values) and Power Regression tables

6
Issues raised Case controlled trial Odds Ratios Confidence intervals, Significant tests (P values) and Power Regression tables

7
Retrospective study design. (looks back in time) Cannot calculate the absolute or true relative risk, attributed risk, but: Cannot calculate the absolute or true relative risk, attributed risk, but: Can calculate the odds ratio Can calculate the odds ratio Control choice is critical. Crucial that the controls are similar to cases with respect to general characteristics. The matching procedure may consider one or several confounding variables (such as age, sex, blood group etc) which may influence the prevalence of the disease being studied. Control choice is critical. Crucial that the controls are similar to cases with respect to general characteristics. The matching procedure may consider one or several confounding variables (such as age, sex, blood group etc) which may influence the prevalence of the disease being studied. Advantages: Advantages: small number of subjects can be used small number of subjects can be used useful for rare diseases useful for rare diseases multiple exposures can be studied and a large number of factors can be examined multiple exposures can be studied and a large number of factors can be examined cheap and easy to do cheap and easy to do Disadvantages: Disadvantages: reliance on recall or records to determine exposure status; reliance on recall or records to determine exposure status; selection of cases and controls can be difficult (population similarities, confounding variables) selection of cases and controls can be difficult (population similarities, confounding variables) not useful for studying the effects of factors/events which are rare. not useful for studying the effects of factors/events which are rare. Does not measure incidence therefore cannot derive risk from this Does not measure incidence therefore cannot derive risk from this Case control study

8
Issues raised Case controlled trial Odds Ratios Confidence intervals, Significant tests (P values) and Power Regression tables

9
A ratio of nonevents to events. If the event rate for a disease is 0.1 (10 per cent), its nonevent rate is 0.9 (90%) and therefore its odds are 9:1 Odds

10
The odds of an experimental patient suffering an adverse event relative to a control patient. Calculated by: exposure rate to risk factor in cases / exposure rate in risk factor in controls It describes the relative importance of a risk factor investigated in a case- control study. The closer the Odds Ratio is to 1, the smaller the difference in effect between the control and experimental group. If more than 1 or less than 1, then the effects of treatment are correspondingly more or less. It is NOT the same as Relative risk (as this requires incidence rates from Cohort) but it is an accurate estimate of relative risk when: the disease group is representative of all cases of the disease, the disease group is representative of all cases of the disease, the control group is similarly representative of the population of people without the disease, the control group is similarly representative of the population of people without the disease, the incidence rate of the disease in the general population is very low. the incidence rate of the disease in the general population is very low. Confidence intervals can be applied to odds ratios Odds Ratio

11
Issues raised Case controlled trial Odds Ratios Confidence intervals, Significant tests (P values) and Power Regression tables

12
You record a number of BP readings and obtain a mean value. Another doctor records some more BP readings on another patient and obtains another mean value. Plot these Means out, there will be a deviation of these means called the standard error of the sample mean (SEM). A large sample of patients provides many Means, the Sample Mean would be a close estimate of the true population Mean, and the standard error would approach close to zero. The Confidence interval provides a measure of the extent to which a sample estimate is likely to differ from this true population. It gives the degree of certainty about the size of difference. A large confidence interval indicates uncertainty, a narrow confidence interval gives greater certainty and higher precision. In a normal distribution, a range of 1.96 standard deviations on either side of the mean will cover 95 per cent of the area under the curve. Confidence interval

13
Confidence intervals become increasingly narrow as the sample size increase. The larger the trial, the narrower the confidence interval and the more definitive the study. 95 per cent confidence interval is +/- 1.96 times the standard error (SE); conversely there is a 5 per cent chance of the true population mean lying outside this range ("<0.05). Study states " The ACE inhibitor group had a 5% (95% confidence interval -1.2 to 12 ) higher survival.. The 95% confidence interval overlaps zero and could be therefore classified as a negative trial, i.e. no difference really exists. However the result lies more to the 12 than –1.2, therefore a more useful conclusion is "an ACE inhibitor is useful in heart failure, but the strength of that inference is weak". Alternatively, if the 95% confidence interval looking at the mean blood glucose levels of two groups ranged from +0.7 to +3.9, as the interval does not cross 0, then we can be 95% sure that the true level lies between +0.7 and +3.9. Confidence Intervals

14
A Significant test is used to decide which hypothesis is accepted and which is rejected. The result is expressed as a P value. They are not the same as confidence intervals. Null hypothesis: A null hypothesis states that there is no difference between the two groups. It is a statement of "no effect" The investigator evaluates an intervention by posing the null hypothesis and then tests it statistically; if it is rejected at a specific level of confidence, e.g. 5 per cent level, then the alternative hypothesis, that there is a difference between groups or a relationship between variables, is accepted with the corresponding degree of confidence, e.g. 95 per cent. Significant test

15
P values are derived from Significant Tests (e.g. Student T Test). The P value looks at how likely the Null hypothesis is true. If the probability is small, then the results of the study are unlikely to occur due to chance (i.e. it is statistically significant). A P value of <0.05 (5 %) or 1 in 20, is usually statistically significant. This level is called the significance level. Significance tests or p values are prone to two types of error:- Type I error or alpha error (false-positive result) occurs if the Null hypothesis is rejected when it is actually true (the effects of a drug are interpreted as being different when they are not). The error is equal to the p value, therefore one can reduce the type I error by making the p value very low (P<0.001) Type I error or alpha error (false-positive result) occurs if the Null hypothesis is rejected when it is actually true (the effects of a drug are interpreted as being different when they are not). The error is equal to the p value, therefore one can reduce the type I error by making the p value very low (P<0.001) Type II error or beta error (false-negative result) occurs if the Null hypothesis is accepted when it is actually false (the effects of the drug are interpreted as being equal when they are actually different). Type II error or beta error (false-negative result) occurs if the Null hypothesis is accepted when it is actually false (the effects of the drug are interpreted as being equal when they are actually different). The P value depends on the power of the study P value

16
The probability that when the study is completed and the data analysed, a statistical significant difference will be detected. ( i.e. the probability that a type 2 error (false negative) will not be made.) The probability that when the study is completed and the data analysed, a statistical significant difference will be detected. ( i.e. the probability that a type 2 error (false negative) will not be made.) If a study has adequate power, it can reliable detect a clinically important difference if one actually exists. Calculating the Power indicates how many patients are needed for the trial, (sample size). In order for the power of the study to be high, In order for the power of the study to be high, the Type I error should be low ( usually 5%) the Type I error should be low ( usually 5%) the type II error at 10% or 20% the type II error at 10% or 20% and that there is a real clinical significant difference. and that there is a real clinical significant difference. These factors affect the number of patients needed to be recruited into the study. Power

17
Issues raised Case controlled trial Odds Ratios Confidence intervals, Significant tests (P values) and Power Regression tables

18
A mathematical equation allowing one variable ( the target variable) to be predicted from another ( the independent variable). multiple regression - a far more complex mathematical formula allowing the target variable to be predicted from two or more independent variables. Regression

Similar presentations

© 2020 SlidePlayer.com Inc.

All rights reserved.

To make this website work, we log user data and share it with processors. To use this website, you must agree to our Privacy Policy, including cookie policy.

Ads by Google