, 23 tweets, 5 min read
My Authors
Read all threads
1/ Okay, new computer is set up. Time to take it for a test run with some #RealtimeScience. I'm going to do some power crunching with NHANES live for a little bit and see how this baby handles...

Feel free to follow along!

[Warning: this thread may have much meandering.]
2/ First, I need to confirm parameters have all the 40,442 in the dataset, and indeed they do when I have everything open-ended...

As always, I caveat that this is very back-of-the-envelope and invite any/everyone to check my work/math and contribute to the conversation. :)
3/
LDL|N|Survival
0-79|6696|88%
80-119|17275|89%
120-159|12167|88%
160-199|3572|86%
200-999|732|86%

There you go -- higher LDL equals less % survival. ;)

Ah... but there's a couple catches when doing the analysis this way...
4/ Let's add one column for the years of follow up for each group...

LDL|N|Survival|Years followup (total)
0-79|6696|88%|7.2
80-119|17275|89%|8
120-159|12167|88%|8.6
160-199|3572|86%|8.9
200-999|732|86%|9.2
5/ Each group has progressively higher and higher average followup time the higher its average LDL-C.

Let's do a poll.

Pretend you follow up Person A in 7.2 years and Person B in 9.2 years and you know nothing else about them. Which is more likely to have died?
6/ The answer is definitely Person B! All else being unknown, checking in on someone at a later period of time means it will be more likely they have died. Put simply, age is a risk factor for death.

But to be sure, followup time is also cut short by death itself as well (ofc)
7/ One way to tackle this is to simply divide % of mortality per year of followup (thanks goes to Dr. Epstein on this solution).

All cause mortality %
/
Years of followup
=
% Mortality per year
8/ But we're not out of the woods yet. If we know age is a risk factor for death, would it matter if we compare the rates of death for younger vs older groups?

All else being unknown, are people who are younger less likely to die each year than people who are older?
9/ Assuming it is indeed a problem, we'll need something referred to as "age parity". Which is to say, we need to try to match up the groups as close in mean age as possible so as not to have the big age offset confounding the results.
10/ As you can see, that's hard to do at this baseline:

LDL|N|Avg Age
0-79 | 6696 | 45
80-119 | 17275 | 45
120-159 | 12167 | 49
160-199 | 3572 | 51
200-999 | 732 | 51

Each grouping of LDL has an average age that is younger (45) in the low and older (51) in the high
11/ The only real solution to this is to try and get to groupings that have age parity. Pref no more than 1 to 2 year differences.

But that can be challenging. I'll see if I can get close by looking at those 50+ w/o statins.
12/
LDL | N | Avg Age | Years followup (total)
0-79 | 1177 | 66 | 6.7
80-119 | 4877 | 65 | 7.6
120-159 | 5200 | 65 | 8.2
160-199 | 1797 | 65 | 8.5
200-999 | 360 | 64 | 8.8

Pretty close, but now low LDL has a bit of offset in avg age at 66 v 64
13/ I'm adding a column to do an offset for each group, starting the baseline age higher until the averages of starting age match between each grouping. This gives me the age parity we need to be fair in this context...
14/
LDL | N | Avg Age
0-79 | 1177 | 66
80-119 | 4651 | 66
120-159 | 4765 | 66
160-199 | 1572 | 66
200-999 | 304 | 66

Now we add the % mortality per year followup...
15/ Pretty interesting...

LDL | N | Avg Age | Mortality per year follow up
0-79 | 1177 | 66 | 4.64%
80-119 | 4651 | 66 | 3.56%
120-159 | 4765 | 66 | 2.9%
160-199 | 1572 | 66 | 3.15%
200-999 | 304 | 66 | 3.1%

Triad* | 387 | 66 | 2.6%
* LDL ≥ 160, HDL ≥ 50, TG ≤ 100
16/ Same but adding HDL ≥ 50 & TG ≤ 100 for everyone:

LDL | N | Avg Age | Mortality per year follow up
0-79 | 362 | 66 | 5.05%
80-119 | 1413 | 66 | 3.7%
120-159 | 1260 | 66 | 2.72%
160-199 | 329 | 66 | 2.93%
200-999 | 44 | 66 | 1.49%
17/ So this is super interesting. Again, more back-of-the-envelop.... I've stratified by their classification, "Diseases of heart (I00-I09, I11, I13, I20-I51)" -- which I'll abrv to "DoH"

There are no other heart or CVD-related codes that I'm aware of...
18/
LDL | N | Avg Age | Mortality per year follow up | DoH
0-79 | 1177 | 66 | 4.64% | 0.95%
80-119 | 4651 | 66 | 3.56% | 0.61%
120-159 | 4765 | 66 | 2.9% | 0.52%
160-199 | 1642 | 66 | 3.05% | 0.53%
200-999 | 320 | 66 | 2.98% | 0.89%
19/ Thus, low LDL of <80 associates greater %/year mortality in the grouping where we got 66 age parity (50ish and up, see above). Yet actually has a higher rate of mortality by Disease of the Heart than the other groupings?

The skeptic in me speculates a few things...
20/ Perhaps this is reverse causality, thus the next step would be to exclude deaths that took place within 5 years of the initial exam.

After that, I can get a better sense on my other questions...
21/ 🤯🤯
LDL | N | Avg Age | Mortality per year follow up | Heart Disease
0-79 | 677 | 66 | 2.75% | 0.55%
80-119 | 3038 | 66 | 2.52% | 0.4%
120-159 | 3512 | 66 | 2.05% | 0.32%
160-199 | 1196 | 66 | 2.25% | 0.34%
200-999 | 217 | 66 | 2.05% | 0.51%
22/ Really, really fascinating. As always, I emphasize these are associations. But certainly very interesting ones at that.

The 120-159 & 200-999 had the lowest %/per year mortality at 2.05%, the 0-79 group the highest at 2.75% -- even after excluding all deaths within 5 years..
23/ What I should do next time is break down the percentages of death associations on a per-grouping basis. I still want to know if low LDL has a much higher rate of death by infection and/or cancer.

Hope this didn't bore anyone to sleep. Cheers!
Missing some Tweet in this thread? You can try to force a refresh.

Enjoying this thread?

Keep Current with Dave Feldman

Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

Twitter may remove this content at anytime, convert it as a PDF, save and print for later use!

Try unrolling a thread yourself!

how to unroll video

1) Follow Thread Reader App on Twitter so you can easily mention us!

2) Go to a Twitter thread (series of Tweets by the same owner) and mention us with a keyword "unroll" @threadreaderapp unroll

You can practice here first or read more on our help page!

Follow Us on Twitter!

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just three indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3.00/month or $30.00/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!