MrMagoo
Senior Member (Voting Rights)
I’m just coming out with random stuff now, isn’t a lot of Canada Scottish as well?
A crumb of a clue on epidemiology
- Thread starter Murph
- Start date
Yeah, it looks to be higher than in the US:
Scottish Canadians
Scottish Americans
Also higher when considering the more general British ancestry:
British Canadians
British Americans
If British ancestry were a risk factor for ME/CFS, then presumably we'd see a larger prevalence in Canada. I don't know if we have any good studies on that.
But if we are considering the Google Trends data for the past 22 years, the search interest in Canada for ME/CFS is only barely higher than in the US (scores are 18 vs. 16), which seems to go against the idea of British propensity for ME/CFS.
One slightly frustrating thing about Google Trends is that the data does not appear to be consistent if downloaded on different days, even if representing the same time span and search term.
The 22 year data for ME/CFS I was using was downloaded on 2026-03-24, and represents the time span of 2004-01-01 to 2004-03-24. I re-downloaded the data for the same time span today, and it is not the same. Of course, I probably shouldn't have included the present date of March 24 within the range of the time span for the original download, as the day wasn't done yet, but the additional few hours shouldn't meaningfully change the results that represent 22 years of data.
Others have commented about the inconsistency elsewhere, with the explanation given that the Trends data is not based on all Google searches, but is instead based on a relatively small sample. On a different day, the search interest for a topic could have been recalculated with a different sample, changing the results.
Thankfully, the values don't change by a huge amount. Here I have plotted the data I used for the previous analyses based on the 22 years of ME/CFS trends data, against the data I have re-downloaded today based on the same time span.
It's highly correlated, so shouldn't change results too much, but I wanted to note this to avoid confusion in case anyone follows the links to the Trends data I provided and sees that what it shows doesn't exactly match what I described in my posts.
The 22 year data for ME/CFS I was using was downloaded on 2026-03-24, and represents the time span of 2004-01-01 to 2004-03-24. I re-downloaded the data for the same time span today, and it is not the same. Of course, I probably shouldn't have included the present date of March 24 within the range of the time span for the original download, as the day wasn't done yet, but the additional few hours shouldn't meaningfully change the results that represent 22 years of data.
Others have commented about the inconsistency elsewhere, with the explanation given that the Trends data is not based on all Google searches, but is instead based on a relatively small sample. On a different day, the search interest for a topic could have been recalculated with a different sample, changing the results.
Thankfully, the values don't change by a huge amount. Here I have plotted the data I used for the previous analyses based on the 22 years of ME/CFS trends data, against the data I have re-downloaded today based on the same time span.
It's highly correlated, so shouldn't change results too much, but I wanted to note this to avoid confusion in case anyone follows the links to the Trends data I provided and sees that what it shows doesn't exactly match what I described in my posts.
Moving right along in our data mining operation...
CDC Wonder is a website which provides access to several interesting public health datasets. I downloaded age-standardized rates of causes of death grouped by state from the Multiple Cause of Death 1999 - 2020 dataset.
This dataset gives the rate of a cause of death in a state whether it was the single "underlying" cause of death, or if it was one of up to 20 additional contributing causes of death listed on the death certificate.
After filtering to only causes which all states had data for, I was left with 598 causes.
I tested Spearman correlation between each cause of death and the Google Trends ME/CFS data (using the same dataset I was using previously: 2004/01/01-2026/03/24), both with and without covariates added for potential confounders.
I added a few more covariates for this analysis that I didn't use previously: proportion of state that has never smoked, sex distribution, healthcare access (proportion who have been to the doctor in the past year) and age. Sources for all covariates:
These are the 15 most significant Spearman correlations for cause of death vs. ME/CFS searches in a state, ranked by p-value controlling for covariates:
Only two causes were significant after strict Bonferroni correction with a 0.05 threshold: muscular dystrophy and alcohol. Poisoning was also below the 0.05 if correcting with FDR.
Alcohol is interesting because it aligns with the results of the previous analysis where we found high correlations with drugs meant for reducing alcohol cravings.
Here is a map showing rates for muscular dystrophy, which had the largest correlation. It doesn't show the distinct grouping of high values seen in both upper corners for both ME/CFS searches and British ancestry.
And a plot of ME/CFS search trends vs. muscular dystrophy rate (note the stats in the corner are univariate Pearson stats):
Edit: I attached a spreadsheet with correlation results for all causes.
CDC Wonder is a website which provides access to several interesting public health datasets. I downloaded age-standardized rates of causes of death grouped by state from the Multiple Cause of Death 1999 - 2020 dataset.
This dataset gives the rate of a cause of death in a state whether it was the single "underlying" cause of death, or if it was one of up to 20 additional contributing causes of death listed on the death certificate.
After filtering to only causes which all states had data for, I was left with 598 causes.
I tested Spearman correlation between each cause of death and the Google Trends ME/CFS data (using the same dataset I was using previously: 2004/01/01-2026/03/24), both with and without covariates added for potential confounders.
I added a few more covariates for this analysis that I didn't use previously: proportion of state that has never smoked, sex distribution, healthcare access (proportion who have been to the doctor in the past year) and age. Sources for all covariates:
These are the 15 most significant Spearman correlations for cause of death vs. ME/CFS searches in a state, ranked by p-value controlling for covariates:
| Cause of death | Spearman R (Univariate) | P value (Univariate) | Spearman R (with covariates) | P value (with covariates) | Spearman P value (with covariates, Bonferroni) | Spearman P value (with covariates, FDR) | |
| 1 | Muscular dystrophy (G71.0) | 0.534 | 5.4E-05 | 0.606 | 2.1E-05 | 0.0125 | 0.0125 |
| 2 | Alcohol, unspecified (T51.9) | 0.508 | 1.4E-04 | 0.579 | 5.9E-05 | 0.0352 | 0.0176 |
| 3 | Poisoning by and exposure to other and unspecified drugs, medicaments and biological substances, undetermined intent (Y14) | 0.387 | 5.0E-03 | 0.558 | 1.2E-04 | 0.0735 | 0.0245 |
| 4 | Perforation of intestine (nontraumatic) (K63.1) | 0.496 | 2.1E-04 | 0.521 | 4.0E-04 | 0.2397 | 0.0599 |
| 5 | Rheumatic heart disease, unspecified (I09.9) | 0.513 | 1.2E-04 | 0.496 | 8.3E-04 | 0.4938 | 0.0854 |
| 6 | Malignant melanoma of skin, unspecified - Malignant neoplasms (C43.9) | 0.616 | 1.5E-06 | 0.495 | 8.6E-04 | 0.5123 | 0.0854 |
| 7 | Other and unspecified narcotics (T40.6) | 0.314 | 2.5E-02 | 0.488 | 1.0E-03 | 0.6181 | 0.0883 |
| 8 | Crohn disease, unspecified (K50.9) | 0.596 | 3.9E-06 | 0.479 | 1.3E-03 | 0.8000 | 0.1000 |
| 9 | Vascular disorder of intestine, unspecified (K55.9) | 0.400 | 3.6E-03 | 0.474 | 1.5E-03 | 0.9046 | 0.1005 |
| 10 | Methadone (T40.3) | 0.496 | 2.1E-04 | 0.467 | 1.8E-03 | 1 | 0.1084 |
| 11 | Motor neuron disease (G12.2) | 0.654 | 1.9E-07 | 0.463 | 2.0E-03 | 1 | 0.1084 |
| 12 | Sequelae of complications of surgical and medical care, not elsewhere classified (T98.3) | 0.437 | 1.4E-03 | 0.456 | 2.4E-03 | 1 | 0.1101 |
| 13 | Acute vascular disorders of intestine (K55.0) | 0.237 | 9.5E-02 | 0.456 | 2.4E-03 | 1 | 0.1101 |
| 14 | Other synthetic narcotics (T40.4) | 0.278 | 4.8E-02 | 0.448 | 3.0E-03 | 1 | 0.1190 |
| 15 | Sequelae of surgical and medical procedures as the cause of abnormal reaction of the patient, or of later complication, without mention of misadventure at the time of the procedure (Y88.3) | 0.408 | 2.9E-03 | 0.444 | 3.3E-03 | 1 | 0.1190 |
Alcohol is interesting because it aligns with the results of the previous analysis where we found high correlations with drugs meant for reducing alcohol cravings.
Here is a map showing rates for muscular dystrophy, which had the largest correlation. It doesn't show the distinct grouping of high values seen in both upper corners for both ME/CFS searches and British ancestry.
And a plot of ME/CFS search trends vs. muscular dystrophy rate (note the stats in the corner are univariate Pearson stats):
Edit: I attached a spreadsheet with correlation results for all causes.
Last edited:
Mortal Machinery
Established Member
Fascinating discussion.
This almost certainly has nothing to do with losartan being "protective" against ME/CFS, and is possibly related to HBP being common in people with diabetes, which as noted earlier is more prevalent in the US South.
But since we are discussing crumbs of clues...
Interesting to see losartan (and the other "sartans") in the negative correlation. I think they may have some effect in lowering or preventing increases in TGF-B? Which if I have understood correctly is one of the few somewhat consistent ME/CFS immune findings.
This almost certainly has nothing to do with losartan being "protective" against ME/CFS, and is possibly related to HBP being common in people with diabetes, which as noted earlier is more prevalent in the US South.
But since we are discussing crumbs of clues...