Initial results

[itchyshin]

@Mike-Garratt

Here is the model with sex

> mod1 <-  rma.mv(yi, V = vi, mod = ~ Sex-1, random = list(~1|Phylogeny, ~1|Species_Latin, ~1|Study, ~1|Effect_ID), R = list(Phylogeny = cor_tree), data = dat, test = "t")
> summary(mod1) 

Multivariate Meta-Analysis Model (k = 113; method: REML)

   logLik   Deviance        AIC        BIC       AICc 
-135.7910   271.5820   283.5820   299.8392   284.3897   

Variance Components:

            estim    sqrt  nlvls  fixed         factor    R 
sigma^2.1  0.1423  0.3772     14     no      Phylogeny  yes 
sigma^2.2  0.0000  0.0001     14     no  Species_Latin   no 
sigma^2.3  0.4436  0.6660     50     no          Study   no 
sigma^2.4  0.2822  0.5312    113     no      Effect_ID   no 

Test for Residual Heterogeneity:
QE(df = 111) = 4478.6286, p-val < .0001

Test of Moderators (coefficients 1:2):
F(df1 = 2, df2 = 111) = 1.0565, p-val = 0.3511

Model Results:

           estimate      se    tval    pval    ci.lb   ci.ub 
SexFemale    0.2100  0.2510  0.8363  0.4048  -0.2875  0.7074    
SexMale      0.3530  0.2638  1.3383  0.1835  -0.1697  0.8758    

---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

And a plot

As you can see and expected, there seems like some effect for males but so much variation.

I have not done the analysis comparing with the opposite sex, which I will do next.

Best and have a good weekend

Shinichi

[itchyshin]

I forgot to mention - interesting Phylogeny seems to account for quite a bit of variation

> round(i2_ml(mod)*100, 2)
        I2_total     I2_Phylogeny I2_Species_Latin         I2_Study     I2_Effect_ID 
           99.95            16.38             0.00            48.22            35.34 

Total heterogeneity is 99.9% so much unaccounted for. Also, study accounts for the half of the variation we see in effect sizes

[Mike-Garratt]

Hi Shinichi,

Sorry, I missed the first part of this message with the data! Super cool thanks!

I am guessing that those two male outliers are the hampster studies, where there is a big reversal in the sex difference in lifespan (e.g. females have a much shorter lifespan than males and male castration reduces this, according to their report).

Let me know if I can help with anything!

Mike

[Mike-Garratt]

Hi Shinichi,

Sorry, I missed the first part of this message with the data! Super cool thanks!

I am guessing that those two male outliers are the hampster studies, where there is a big reversal in the sex difference in lifespan (e.g. females have a much shorter lifespan than males and male castration reduces this, according to their report).

Let me know if I can help with anything!

Mike

[itchyshin]

@Mike-Garratt

OK - I did read the Word doc and tested these moderators already but some of them seemed to have levels with so few sample sizes (e.g. n = 2; Type of sterilization) so I was not sure how meaningful they are. Of course, we need to balance what is good statistically and what we want to test.

So it will be good to catch up with you to chat after I explore a bit more. I will do a comparison between treatments and the opposite sex as well.

I will be away between 14 and 19th so it will be good if we catch up after that.

Best

Shinichi

[Mike-Garratt]

Thanks Shinichi. I get your point about the type of sterilization moderator. This question may be most relevant in females, where there are several different manipulations used in more than a few studies. We also have a gonad_removed moderator which should have more statistical power and is basically a more simple classification of the most important difference between the treatments - whether gonads were removed or not. This will influence whether sex hormone production is removed or not, which is expected to influence aging in males and females in different ways.

Sounds good - I look forward to chatting when you get back. I will work on that age at treatment moderator in the mean time.

Cheers

Mike

[Mike-Garratt]

Hey Shinichi, I hope all is going well and you had a good time away. Just wondering if you wanted to set up a time to chat about the meta-analysis. Just let me know when you are free, there is no rush, I can imagine you are mega-busy!

Speak soon

Mike

[itchyshin]

Sure @Mike-Garratt - how about my 2 pm on Tuesday (next week - 27th April - your 4 pm - I think). If it is fine, you can send me a Zoom link? I will try to do a bit more before we meet then.

[itchyshin]

@Mike-Garratt

My note (list) for our meeting

• What should we do with "outliers" - Hamster data
• Can you compress the "Environment" variable
• No overall significance pretty much everything - due to very high heterogeneity
• Here is the results of the opposite sex analysis (compared with the above) - they seem to narrow the gap but don't we need to know what the gap in longevity between the sexes to start off with?

Multivariate Meta-Analysis Model (k = 68; method: REML)

  logLik  Deviance       AIC       BIC      AICc 
-80.0992  160.1984  172.1984  185.3363  173.6221   

Variance Components:

            estim    sqrt  nlvls  fixed         factor    R 
sigma^2.1  0.2725  0.5220     10     no      Phylogeny  yes 
sigma^2.2  0.0000  0.0001     10     no  Species_Latin   no 
sigma^2.3  0.6166  0.7852     25     no          Study   no 
sigma^2.4  0.2584  0.5083     68     no      Effect_ID   no 

Test for Residual Heterogeneity:
QE(df = 66) = 3011.6093, p-val < .0001

Test of Moderators (coefficients 1:2):
F(df1 = 2, df2 = 66) = 0.9377, p-val = 0.3967

Model Results:

           estimate      se    tval    pval    ci.lb   ci.ub 
SexFemale    0.0268  0.3743  0.0716  0.9431  -0.7205  0.7742    
SexMale      0.2340  0.3737  0.6262  0.5333  -0.5121  0.9801   

[Mike-Garratt]

Hey Shinichi,

Thanks for sending this through, I'll get to do some prep before our meeting. I was thinking that for the opposite sex analysis we look at the difference in lifespan between control males and control females (which we have for all the data where there is an opposite sex survival column filled in), and then compare this to the difference in lifespan between the sterilized animals and the opposite sex. So for example, we would have control male versus control female survival (lets say a difference in 5 years), then the difference in lifespan between the castrated male and the female (lets say 2 years). we would expect the second number to be smaller if castration reduces the sex difference in survival (as predicted).

The other thing I was wondering if we should check is whether the mean/median data gives a different standardized difference compared to the survival rate data, which may mask some of the effects? I think that you said the survival rate data showed a massive effect when you first looked.

I'll try to send through any other talking points before our meeting, although its a long weekend here.

Hope all is going well,

Mike

[itchyshin]

Hi, Mike

Ahh! - of course - I should have realised this - I will do the analysis you suggested - to do this properly is a bit tricky but leave this to me.

I already checked the difference between these two types: survival rate and mean/median - no sig. differences

Thanks

Shinichi

[Mike-Garratt]

[Mike-Garratt]

[Mike-Garratt]

Pasted above are the plots of the change in survival (sterilised animal value/intact animals value) with sterilisation relative to the sex difference in survival (opposite sex survival/intact animal of the sex in question). This type of analysis, where the same denominator is used in two ratios, will produce spurious results, although I think that the strength of effect in males is not attributed to this. Also the relationship is highly significant in males but not significant in females (and there is a significant interaction), which also points that there could be something real going on. Although note that the X axis is different in the two plots.

We can also test whether the variance in survival between males and females goes down with sterilization, which I think will get around this problem of spurious correlations. I am trying to play around with this but cant find the best way to do this currently.

Thanks for the chat today, I'll follow up with Jeff about the sexual maturity data.

Cheers

Mike

[Mike-Garratt]

Hey Shinichi,

Just a quick note, I realized that I had entered data for mortality rate rather than survival rate for two studies (Parker et al 2013 and Wilson et al 2019). This has now been corrected in the final extracted data file (I have also corrected the study name for Parker et al).It probably wont make a difference on the results because the effect was small, but thought I should let you know.

No worries about not getting to the other analysis. I am trying to finish this new age at treatment moderator, so hopefully that one will be ready by the time to get chance to come back to it.

Hope all is well,

Mike

[Mike-Garratt]

Hi Shinichi,

I've now added some revised moderators to the sterilization dataset based on our discussions.

An ordinal variable for age at sterilization. I think the best way to do this is Birth (1), Prepuberty (2), puberty (3) and adult (4). I did try and seperate out the age at adulthood but it is probably too messy, since we have this information for some studies but not others.

One thing that limits us from having nearly complete information for this are the studies on cats and dogs. It seems that these surgeries are mostly conducted at 6-12 months of age, although there is definitely variably and sometimes it is done earlier and sometimes later. We do have a couple of studies where we have the data for age at sterilization for these species, so it is probably best to just leave this.

Ive also added another variable to try to condense the “environment” variable, which is whether animals lived in a wild or semi-wild environment, which is whether where they were protected from predation and supplemented with food or not. This is the best way I can think of allocating the environment variables into meaningful smaller groups.

Let me know if you want any further information about the other sex-differences analysis we talked about a couple of weeks ago.

Hope all is well,

Mike

[itchyshin]

@Mike-Garratt

Thanks for this - sounds great. I have to do teaching preparation this week so I may not get this till next week. I will try my best. Realistically speaking, let's say I get something to you by the 21st of May, if not earlier. Hope it is OK.

Best

Shinichi

[Mike-Garratt]

No problems, I totally understand! Let me know if you have any questions when you get chance to get to it.

Cheers

Mike