Closed ld-archer closed 3 years ago
Cancer: Cancer rates from the model are much higher than ELSA by wave 8 (20% vs 16%). To improve the cancer model:
Update
Made big improvements in the chronic disease models, and risk behaviours look slightly better (both still WIP).
variable | fem_mean_wave3 | elsa_mean_wave3 | p_value_wave3 | fem_mean_wave4 | elsa_mean_wave4 | p_value_wave4 | fem_mean_wave5 | elsa_mean_wave5 | p_value_wave5 | fem_mean_wave6 | elsa_mean_wave6 | p_value_wave6 | fem_mean_wave7 | elsa_mean_wave7 | p_value_wave7 | fem_mean_wave8 | elsa_mean_wave8 | p_value_wave8 |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Any ADLs | 0.20446 | 0.21265 | 0.25128 | 0.20819 | 0.22858 | 0.00916 | 0.21407 | 0.23353 | 0.01708 | 0.22762 | 0.23014 | 0.76752 | 0.2423 | 0.21925 | 0.01056 | 0.25394 | 0.22498 | 0.003 |
Any IADLs | 0.22208 | 0.22211 | 0.99636 | 0.23023 | 0.24747 | 0.03216 | 0.238 | 0.2548 | 0.04566 | 0.25351 | 0.26006 | 0.46049 | 0.27101 | 0.25179 | 0.04184 | 0.29071 | 0.26585 | 0.01591 |
Cancer ever | 0.10008 | 0.08634 | 0.00535 | 0.11492 | 0.0972 | 0.00144 | 0.12968 | 0.11917 | 0.094 | 0.14571 | 0.12381 | 0.00108 | 0.16072 | 0.1466 | 0.06695 | 0.17607 | 0.16397 | 0.16124 |
Diabetes ever | 0.11092 | 0.0937 | 0.00077 | 0.12531 | 0.11321 | 0.04139 | 0.13843 | 0.12803 | 0.10811 | 0.15527 | 0.13755 | 0.01134 | 0.16736 | 0.15406 | 0.09058 | 0.17651 | 0.15097 | 0.00229 |
Heart disease ever | 0.21407 | 0.17494 | 0 | 0.23299 | 0.1983 | 0 | 0.25353 | 0.22183 | 8E-05 | 0.27484 | 0.22953 | 0 | 0.2967 | 0.25602 | 2E-05 | 0.31941 | 0.28985 | 0.00524 |
Hypertension ever | 0.47905 | 0.4483 | 0.0004 | 0.50866 | 0.48162 | 0.00377 | 0.53503 | 0.50639 | 0.00303 | 0.55888 | 0.50793 | 0 | 0.58066 | 0.53135 | 1E-05 | 0.602 | 0.5384 | 0 |
Lung disease ever | 0.09032 | 0.07187 | 5E-05 | 0.09938 | 0.07859 | 4E-05 | 0.10795 | 0.08631 | 8E-05 | 0.11846 | 0.08619 | 0 | 0.12791 | 0.09168 | 0 | 0.13641 | 0.08975 | 0 |
Psychological problems ever | 0.09947 | 0.08478 | 0.00268 | 0.11003 | 0.09286 | 0.00164 | 0.12086 | 0.09286 | 0 | 0.13097 | 0.09792 | 0 | 0.14043 | 0.10107 | 0 | 0.151 | 0.11419 | 0 |
Stroke ever | 0.05912 | 0.05037 | 0.02287 | 0.06475 | 0.06373 | 0.82334 | 0.07223 | 0.06743 | 0.32293 | 0.07805 | 0.07311 | 0.3497 | 0.08433 | 0.0833 | 0.86354 | 0.09276 | 0.08899 | 0.5701 |
Something that would benefit the smoking and drinking models is an indicator of intensity. We've included this before with the smokef (# cigs) and drink_stat vars, but neither of these worked particularly well. Now might be a good time to give that a try, going to spin this comment off into a new issue.
Models are now pretty reasonable, and logic to calculate them is in the debug document.
Using the T-tests as a benchmark for improvement, we need to improve the transition models for some chronic diseases.
We will focus on Cancer, Diabetes, Heart Disease, and Hypertension first, as these are some of the most important chronic diseases, and some are highly influential in other chronic diseases.
Here are the T-tests as a starting point:
Ideas: So far, I have made some reasonable improvements just by tweaking the transition models. Improving the logbmi model made probably the biggest difference of anything for a couple of variables. I will continue to do this, but also look into some other variables from ELSA we can include. Examples being high cholesterol and hip fracture as 2 variables that have made a noticeable improvement.
This issue continues the work started in #41, #45, #48 & #49.