Skip to 0 minutes and 14 seconds So let me give you an example of this. Some authors Cook and Sackett took a study and looked at it and divided patients into treatment groups who had moderate hypertension and then their event rates and compared it to patients with moderate or to mild hypertension and compare these versus placebo. So they went in and compared the RR, the RRR, ARR and NNT for these two groups. Now you can see that the patients with moderate hypertension had an RR and RRR as 60% and 40%, but so did the mild hypertension patients. But then what happens when you look at the ARR in the NNT so you see those numbers are vastly different because this takes the baseline risk.
Skip to 1 minute and 0 seconds So you could imagine that patients with moderate hypertension are being at a higher risk, so that’s why you see the number NNT of 13 whereas mild hypertension has an NNT of 167. Well we’ll talk about this a little bit more but when you’re looking at NNT in a trial you want a smaller number because you want to treat fewer patients to see one additional benefit. And then the NNH you want those to be larger numbers because you want to treat more patients before the emergence of that adverse effect. So you can see how this data applies to patient care.
Skip to 1 minute and 35 seconds Patients with mild hypertension often aren’t treated because of the potential lack of benefit of meaning 167 patients before you see that additional effect. With moderate hypertension you only have to treat thirteen before you see that ditional effect. So you can see how in studies, authors reporting RR and RRR can often be misleading and so as an evaluator it’s usually good to calculate ARR and NNT to be able to be certain that this type of situation isn’t occurring. So we can’t really talk about measures of Association equations or examples unless we talk about the confidence intervals that will be used for them. Now measures of Association confidence intervals are going to be listed at the bottom but there’s also another type.
Skip to 2 minutes and 25 seconds Reason I’m going over them here as we’ll need the when we get to the non-inferiority section in the next session. So confidence intervals are an estimate from the sample of where the true value of the population would lie. For a mean data regular CI this would be the true mean. For measures of Association it would be the value because that’s nominal data. So you can see that the most common and this is primarily for both superior and non-inferiority trials that it’s 90 to 95% and so you’ll see that the mean data ones or the regular ones that you want the range to not include one .
Skip to 3 minutes and 7 seconds Well or zero excuse me zero means no effect so that should be fairly simple to understand. So these are displayed as and the second bullet is an example. Five point one is the average that is seen in the sample and the 95% confidence interval is what you would see potentially see in the population. Now remember if it touches across a zero that means there’s a chance that you would see no effect in the population based on the sample so you can see in this one that that confidence interval is not clinically relevant because it includes the number zero. You also want to as the as you evaluate these articles.
Skip to 3 minutes and 45 seconds You want to look at the width or the range of the CI. Because the narrow CI implies that the date the sample data was more precise and it more can accurately predict what’s in the population whereas a wider CI provides with less confidence. Obviously confidence intervals. So you would maybe want to not be a circle or you wouldn’t be in a certain that what you saw in the population was actually true based on a sample that wasn’t very reliable.
Skip to 4 minutes and 16 seconds Now along with the CI it’s often a good idea to also have a p-value because you want to make sure if the confidence interval is significant so is the p-value and if there if one is significant one is not there’s a problem somewhere with the internal validity of the study. So you want to make sure that they provide both information for you. So now how do we look at the measures of Association CIs. Well this case you remember it’s an HR and OR and RR. There are all ratios. So instead of the number outside of the confidence in the role being just a mean this time it’s a ratio. So it’s something to 1.
Skip to 4 minutes and 56 seconds 1 being the standard that it’s being compared to. Now most of the outcomes you’re looking at in measures of associations are negative or bad outcomes. So normally you want that number outside of the confidence interval to be less than 1 because that would infer there was less of a risk compared to the standard which would be 1 and it’s greater than 1 that means there’s an increased risk.
Skip to 5 minutes and 20 seconds So here’s an example of an MOA 1 the way you tell the difference that I’m sorry I don’t have it here is that it has to have an RR OR or HR somewhere in the explanation for you to see that it’s a measures and out a general CI But you read them the same way. 0.75 is what was seen in the in the sample and the 0.5 - 0.9 is what would be predicted to be seen in the population. So you can see in this example the confidence interval does not touch across one.
Skip to 5 minutes and 52 seconds So this one potentially could be clinical and relevant but it’s depending on what we’re looking for and again we’d want to see a p-value to confirm these results. So if the CI includes one as I mentioned there’s no difference between treated and non-treaty groups so in that case there’s a possibility in the population you’d see a one-to-one ratio which means you’d see no additional benefit in the treatment compared to the control. And just with general s just as with general CIs, the narrower the CI implies more precise data and the wider it is. It means the data is less reliable.
Baseline Risk and Confidence Intervals
Prof. Mary Ferrill demonstrates why baseline risk matters with an example for moderate hypertension (HTN) and mild hypertension.
She compares the RR, the RRR, ARR, and NNT for the two groups in this study.
Also, she emphasizes that we need a smaller number on NNT because you want to treat fewer patients to see one additional benefit. On the contrary, we want a larger number on NNH because you want to treat more patients before the emergence of that adverse effect.
Then it comes to the confidence intervals (CI). We need to know its definition, common range and types in this section. Dealing with the mean data or regular CI, if the range includes zero, it is possible that no effect occurred.
Ultimately, we can learn how to interpret CIs of Measures of Association (MOA). What does narrow CI mean? Does it mean the data is reliable? Please share your thoughts below.
If you have any questions about this part, please don’t hesitate to leave them below.