Great write-up with a lot of important points, but:
>Cremieux not only took out the HSV infection line, but he also changed the y-axis from cumulative risk of dementia to cumulative incidence of dementia. The original graph is a log-rank test, while Cremieux’s version is, uh, nonsense. Presumably he thought the original meant something like “percentage of people who got dementia”, but it does not. It’s much closer to a hazard ratio, and therefore more difficult to understand and less memeable for the average Tweeter.
I'm not sure you're any more correct than the original tweet. There's no difference between cumulative incidence and cumulative risk (the original article calls it both of those). What does it mean for a graph to be "a log-rank test" (which is a statistical test, not a graph)? I'd normally call a plot like this a Kaplan-Meier survival curve, except that this one isn't plotting survival but presumably its inverse. But it's also totally unclear what the units of the original graph are. They're labelled 'cumulative risk', but that should be a number less than 1. It's not a hazard ratio like you've described it, since that would require a ratio of two groups and is (by assumption) not a function of time. It also appears that the tweet is wrong to have just assumed 2 means 0.02 since the paper goes on to say:
> Subjects with HSV infections taking anti-herpetic medications (N = 7215) were compared to those patients not taking them (N = 1147). In the subgroup with anti-herpetic medications treatment, 419 (5.80%) developed dementia in the longitudinal follow-up within 10 years. In the subgroup without anti-herpetic medications treatment, 325 (28.33%) developed dementia in the same follow-up period.
It looks like the risk at 10 years in the HSV and no-treatment group should be at 28% but is plotted instead at about 1.5. So I think the plot is nonsense but that the nonsense goes back not to twitter but to the original article. The article also gives no indication about loss to follow-up and censoring from that, which is a bit of an omission.
3. People with Epstein-Barr get dementia, and antiherpetic medications help prevent dementia by preventing the replication of EBV.
I don't know if the first part is true, but I wouldn't be surprised since we already know that EBV seroconversion can be followed by terrible things happening in your brain. Unfortunately for the second part, Medscape says "Acyclovir and ganciclovir may reduce EBV shedding, but are ineffective clinically." Maybe every little bit helps though...
"Going to the doctor for Herpes 3 times in a year" then becomes a symptom of having an immune system so dysfunctional that it's allowing EBV to run amok.
Great write-up with a lot of important points, but:
>Cremieux not only took out the HSV infection line, but he also changed the y-axis from cumulative risk of dementia to cumulative incidence of dementia. The original graph is a log-rank test, while Cremieux’s version is, uh, nonsense. Presumably he thought the original meant something like “percentage of people who got dementia”, but it does not. It’s much closer to a hazard ratio, and therefore more difficult to understand and less memeable for the average Tweeter.
I'm not sure you're any more correct than the original tweet. There's no difference between cumulative incidence and cumulative risk (the original article calls it both of those). What does it mean for a graph to be "a log-rank test" (which is a statistical test, not a graph)? I'd normally call a plot like this a Kaplan-Meier survival curve, except that this one isn't plotting survival but presumably its inverse. But it's also totally unclear what the units of the original graph are. They're labelled 'cumulative risk', but that should be a number less than 1. It's not a hazard ratio like you've described it, since that would require a ratio of two groups and is (by assumption) not a function of time. It also appears that the tweet is wrong to have just assumed 2 means 0.02 since the paper goes on to say:
> Subjects with HSV infections taking anti-herpetic medications (N = 7215) were compared to those patients not taking them (N = 1147). In the subgroup with anti-herpetic medications treatment, 419 (5.80%) developed dementia in the longitudinal follow-up within 10 years. In the subgroup without anti-herpetic medications treatment, 325 (28.33%) developed dementia in the same follow-up period.
It looks like the risk at 10 years in the HSV and no-treatment group should be at 28% but is plotted instead at about 1.5. So I think the plot is nonsense but that the nonsense goes back not to twitter but to the original article. The article also gives no indication about loss to follow-up and censoring from that, which is a bit of an omission.
How about:
3. People with Epstein-Barr get dementia, and antiherpetic medications help prevent dementia by preventing the replication of EBV.
I don't know if the first part is true, but I wouldn't be surprised since we already know that EBV seroconversion can be followed by terrible things happening in your brain. Unfortunately for the second part, Medscape says "Acyclovir and ganciclovir may reduce EBV shedding, but are ineffective clinically." Maybe every little bit helps though...
"Going to the doctor for Herpes 3 times in a year" then becomes a symptom of having an immune system so dysfunctional that it's allowing EBV to run amok.