In lwjohnst86/acdcourse: Course on Analyzing Cohort Datasets in R
Cohorts are important scientific sources of health and wellness information.
Because of this, how results are presented needs to be carefully considered. The
medium of presentation, be it plots or tables, can impact how the findings are
seen and consumed. In this chapter we will cover some tips and ways of
presenting cohort findings.
Presenting cohort findings is tricky, be careful
Since cohort studies focus on health and disease, the way results are presented
can directly impact people's lives and health, so care needs to be taken!
Describe results cautiously, carefully
- STROBE: "Give a cautious overall interpretation of results"
- Emphasize tangible health impact
- Be clear and understandable
- Get feedback from healthcare practitioners
- Both "non-significant" and "significant" results important
- Show magnitude and uncertainty over "statistical significance"
In observational cohort studies, you need to use a healthy dose of caution when
interpreting and presenting results, as indicated in the STROBE best practices.
STROBE, if you recall, was designed to make cohort studies more rigorous and
standardized.
Emphasize more tangible health impacts when presenting results and try to be as
clear and understandable as possible. Your audience or readers will likely be
public health professionals or clinicians who aren't trained in interpreting
complex models, so be simple and concise.
Keep in mind that both non-significant and significant findings are important
from a health standpoint. Show all results, emphasizing the magnitude of
association and the uncertainty of that association.
Causal reasoning in epidemiology
- Growing field in epidemiology
- Use caution with causal language or interpretations
- Using stronger causal language depends on magnitude and temporal response
- E.g. with smoking and lung cancer
Oftentimes, observational findings are interpreted causally. While the field of
causal reasoning in epidemiology has expanded recently, it is still new, so use
caution with causal language.
Unfortunately, humans often interpret associations as causal, so as the
researcher you need to be extremely careful when presenting findings from
observational research.
There are however times when causal language is justified, for example with the
observation of cigarette smoking and lung cancer. But usually it isn't
completely clear whether a causal relationship exists. Keep these in mind when
you present cohort study results.
Multiple models as lists, tidying with map
unadjusted_models_list <- list(
glmer(got_cvd ~ total_cholesterol_scaled + (1 | subject_id),
family = binomial, data = tidied_framingham),
glmer(got_cvd ~ body_mass_index_scaled + (1 | subject_id),
family = binomial, data = tidied_framingham)
)
# Same done for adjusted_models_list
library(purrr)
tidied_unadjusted_models_list <- unadjusted_models_list %>%
map(~tidy(.x, conf.int = TRUE, exponentiate = TRUE) %>%
select(effect, term, estimate, conf.low, conf.high))
Given you'll likely be working with multiple models, my suggestion is put your
models into a single list, since working with lists is very easy in R thanks to
the purrr package.
Here, I've ran two models using what we learned from the previous chapter and
stored them as a list into the object unadjusted models list. I did the same
thing for adjusted models by storing them as a list, but haven't shown the code.
Then, to apply the tidy function to each of those models, we use the map
function from the purrr package. Map takes two arguments. The list object and
the function we want to use. We use the function with the tilde symbol
beforehand and refer to the model object by dot x.
Using map leverages R's powerful functional programming strengths. purrr
provides a wonderful and consistent interface for using this functional
programming.
Contents are multiple models stored in sequence
unadjusted_models_list
# adjusted_models_list
[[1]]
# A tibble: 3 x 5
effect term estimate conf.low conf.high
<chr> <chr> <dbl> <dbl> <dbl>
1 fixed (Intercept) 0.00000273 0.000000334 0.0000223
2 fixed total_cholesterol_scaled 1.10 0.319 3.78
3 ran_pars sd__(Intercept) 56.5 NA NA
[[2]]
# A tibble: 3 x 5
effect term estimate conf.low conf.high
<chr> <chr> <dbl> <dbl> <dbl>
1 fixed (Intercept) 0.00000248 0.000000289 0.0000213
2 fixed body_mass_index_scaled 1.62 0.410 6.38
3 ran_pars sd__(Intercept) 56.8 NA NA
The contents are a list of two model results. These results are for the
unadjusted models. The adjusted models list looks the same, except there are
additional predictors.
Plotting or creating tables of these results would be better done using a single
dataframe, so we'll need to combine them together. But before we do, we'll need
to add some details to distinguish each model.
Use map to add more model details
map(unadjusted_models_list, ~mutate(.x, model = "Unadjusted"))
[[1]]
# A tibble: 3 x 6
effect term estimate conf.low conf.high model
<chr> <chr> <dbl> <dbl> <dbl> <chr>
1 fixed (Intercept) 0.00000273 0.000000334 2.23e-5 Unadjust…
2 fixed total_cholesterol_scal… 1.10 0.319 3.78e+0 Unadjust…
3 ran_pars sd__(Intercept) 56.5 NA NA Unadjust…
[[2]]
# A tibble: 3 x 6
effect term estimate conf.low conf.high model
<chr> <chr> <dbl> <dbl> <dbl> <chr>
1 fixed (Intercept) 0.00000248 0.000000289 0.0000213 Unadjust…
2 fixed body_mass_index_scaled 1.62 0.410 6.38 Unadjust…
3 ran_pars sd__(Intercept) 56.8 NA NA Unadjust…
Let's use map to add more information to the model objects in the list. We'll
add a tag to each model item in the list to indicate that the models are
unadjusted.
The model column is now added to each model in the list.
Use bind_rows to convert list into data frame
map(unadjusted_models_list, ~mutate(.x, model = "Unadjusted")) %>%
bind_rows()
# A tibble: 6 x 6
effect term estimate conf.low conf.high model
<chr> <chr> <dbl> <dbl> <dbl> <chr>
1 fixed (Intercept) 0.00000273 3.34e-7 2.23e-5 Unadjus…
2 fixed total_cholesterol… 1.10 3.19e-1 3.78e+0 Unadjus…
3 ran_pars sd__(Intercept) 56.5 NA NA Unadjus…
4 fixed (Intercept) 0.00000248 2.89e-7 2.13e-5 Unadjus…
5 fixed body_mass_index_s… 1.62 4.10e-1 6.38e+0 Unadjus…
6 ran_pars sd__(Intercept) 56.8 NA NA Unadjus…
Then, we can convert the list of models into a single dataframe of the results.
We do this using dplyr's bind_rows() function to stack the list
dataframes into one.
Bind multiple lists and continue piping
bind_rows(
map(unadjusted_models_list, ~mutate(.x, model = "Unadjusted")),
map(adjusted_models_list, ~mutate(.x, model = "Adjusted"))
) %>%
mutate(outcome = "got_cvd")
# A tibble: 14 x 7
effect term estimate conf.low conf.high model outcome
<chr> <chr> <dbl> <dbl> <dbl> <chr> <chr>
1 fixed (Intercept) 0.00000273 0.000000334 2.23e-5 Unadjust… got_cvd
2 fixed total_cholesterol_scal… 1.10 0.319 3.78e+0 Unadjust… got_cvd
3 ran_pars sd__(Intercept) 56.5 NA NA Unadjust… got_cvd
4 fixed (Intercept) 0.00000248 0.000000289 2.13e-5 Unadjust… got_cvd
5 fixed body_mass_index_scaled 1.62 0.410 6.38e+0 Unadjust… got_cvd
6 ran_pars sd__(Intercept) 56.8 NA NA Unadjust… got_cvd
7 fixed (Intercept) 0.00000532 0.000000503 5.61e-5 Adjusted got_cvd
8 fixed total_cholesterol_scal… 1.21 0.347 4.22e+0 Adjusted got_cvd
9 fixed sexWoman 0.272 0.0143 5.15e+0 Adjusted got_cvd
10 ran_pars sd__(Intercept) 55.5 NA NA Adjusted got_cvd
Let's do the same thing with the list of adjusted models, but instead, we'll
bind rows of both unadjusted and adjusted models. This puts all model items into
a single dataframe. Lastly, let's add information about the outcome. We now have
a single model dataframe that we can use to create plots and tables.
Let's practice wrangling!
Now you know how to wrangling model results. Let's practice!
lwjohnst86/acdcourse documentation built on June 18, 2019, 8:26 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Cohorts are important scientific sources of health and wellness information. Because of this, how results are presented needs to be carefully considered. The medium of presentation, be it plots or tables, can impact how the findings are seen and consumed. In this chapter we will cover some tips and ways of presenting cohort findings.
Presenting cohort findings is tricky, be careful
Since cohort studies focus on health and disease, the way results are presented can directly impact people's lives and health, so care needs to be taken!
Describe results cautiously, carefully
- STROBE: "Give a cautious overall interpretation of results"
- Emphasize tangible health impact
- Be clear and understandable
- Get feedback from healthcare practitioners
- Both "non-significant" and "significant" results important
- Show magnitude and uncertainty over "statistical significance"
In observational cohort studies, you need to use a healthy dose of caution when interpreting and presenting results, as indicated in the STROBE best practices. STROBE, if you recall, was designed to make cohort studies more rigorous and standardized.
Emphasize more tangible health impacts when presenting results and try to be as clear and understandable as possible. Your audience or readers will likely be public health professionals or clinicians who aren't trained in interpreting complex models, so be simple and concise.
Keep in mind that both non-significant and significant findings are important from a health standpoint. Show all results, emphasizing the magnitude of association and the uncertainty of that association.
Causal reasoning in epidemiology
- Growing field in epidemiology
- Use caution with causal language or interpretations
- Using stronger causal language depends on magnitude and temporal response
- E.g. with smoking and lung cancer
Oftentimes, observational findings are interpreted causally. While the field of causal reasoning in epidemiology has expanded recently, it is still new, so use caution with causal language.
Unfortunately, humans often interpret associations as causal, so as the researcher you need to be extremely careful when presenting findings from observational research.
There are however times when causal language is justified, for example with the observation of cigarette smoking and lung cancer. But usually it isn't completely clear whether a causal relationship exists. Keep these in mind when you present cohort study results.
Multiple models as lists, tidying with map
unadjusted_models_list <- list( glmer(got_cvd ~ total_cholesterol_scaled + (1 | subject_id), family = binomial, data = tidied_framingham), glmer(got_cvd ~ body_mass_index_scaled + (1 | subject_id), family = binomial, data = tidied_framingham) ) # Same done for adjusted_models_list
library(purrr) tidied_unadjusted_models_list <- unadjusted_models_list %>% map(~tidy(.x, conf.int = TRUE, exponentiate = TRUE) %>% select(effect, term, estimate, conf.low, conf.high))
Given you'll likely be working with multiple models, my suggestion is put your models into a single list, since working with lists is very easy in R thanks to the purrr package.
Here, I've ran two models using what we learned from the previous chapter and stored them as a list into the object unadjusted models list. I did the same thing for adjusted models by storing them as a list, but haven't shown the code.
Then, to apply the tidy function to each of those models, we use the map function from the purrr package. Map takes two arguments. The list object and the function we want to use. We use the function with the tilde symbol beforehand and refer to the model object by dot x.
Using map leverages R's powerful functional programming strengths. purrr provides a wonderful and consistent interface for using this functional programming.
Contents are multiple models stored in sequence
unadjusted_models_list
# adjusted_models_list
[[1]] # A tibble: 3 x 5 effect term estimate conf.low conf.high <chr> <chr> <dbl> <dbl> <dbl> 1 fixed (Intercept) 0.00000273 0.000000334 0.0000223 2 fixed total_cholesterol_scaled 1.10 0.319 3.78 3 ran_pars sd__(Intercept) 56.5 NA NA [[2]] # A tibble: 3 x 5 effect term estimate conf.low conf.high <chr> <chr> <dbl> <dbl> <dbl> 1 fixed (Intercept) 0.00000248 0.000000289 0.0000213 2 fixed body_mass_index_scaled 1.62 0.410 6.38 3 ran_pars sd__(Intercept) 56.8 NA NA
The contents are a list of two model results. These results are for the unadjusted models. The adjusted models list looks the same, except there are additional predictors.
Plotting or creating tables of these results would be better done using a single dataframe, so we'll need to combine them together. But before we do, we'll need to add some details to distinguish each model.
Use map to add more model details
map(unadjusted_models_list, ~mutate(.x, model = "Unadjusted"))
[[1]] # A tibble: 3 x 6 effect term estimate conf.low conf.high model <chr> <chr> <dbl> <dbl> <dbl> <chr> 1 fixed (Intercept) 0.00000273 0.000000334 2.23e-5 Unadjust… 2 fixed total_cholesterol_scal… 1.10 0.319 3.78e+0 Unadjust… 3 ran_pars sd__(Intercept) 56.5 NA NA Unadjust… [[2]] # A tibble: 3 x 6 effect term estimate conf.low conf.high model <chr> <chr> <dbl> <dbl> <dbl> <chr> 1 fixed (Intercept) 0.00000248 0.000000289 0.0000213 Unadjust… 2 fixed body_mass_index_scaled 1.62 0.410 6.38 Unadjust… 3 ran_pars sd__(Intercept) 56.8 NA NA Unadjust…
Let's use map to add more information to the model objects in the list. We'll add a tag to each model item in the list to indicate that the models are unadjusted.
The model column is now added to each model in the list.
Use bind_rows to convert list into data frame
map(unadjusted_models_list, ~mutate(.x, model = "Unadjusted")) %>% bind_rows()
# A tibble: 6 x 6 effect term estimate conf.low conf.high model <chr> <chr> <dbl> <dbl> <dbl> <chr> 1 fixed (Intercept) 0.00000273 3.34e-7 2.23e-5 Unadjus… 2 fixed total_cholesterol… 1.10 3.19e-1 3.78e+0 Unadjus… 3 ran_pars sd__(Intercept) 56.5 NA NA Unadjus… 4 fixed (Intercept) 0.00000248 2.89e-7 2.13e-5 Unadjus… 5 fixed body_mass_index_s… 1.62 4.10e-1 6.38e+0 Unadjus… 6 ran_pars sd__(Intercept) 56.8 NA NA Unadjus…
Then, we can convert the list of models into a single dataframe of the results.
We do this using dplyr's bind_rows() function to stack the list
dataframes into one.
Bind multiple lists and continue piping
bind_rows( map(unadjusted_models_list, ~mutate(.x, model = "Unadjusted")), map(adjusted_models_list, ~mutate(.x, model = "Adjusted")) ) %>% mutate(outcome = "got_cvd")
# A tibble: 14 x 7 effect term estimate conf.low conf.high model outcome <chr> <chr> <dbl> <dbl> <dbl> <chr> <chr> 1 fixed (Intercept) 0.00000273 0.000000334 2.23e-5 Unadjust… got_cvd 2 fixed total_cholesterol_scal… 1.10 0.319 3.78e+0 Unadjust… got_cvd 3 ran_pars sd__(Intercept) 56.5 NA NA Unadjust… got_cvd 4 fixed (Intercept) 0.00000248 0.000000289 2.13e-5 Unadjust… got_cvd 5 fixed body_mass_index_scaled 1.62 0.410 6.38e+0 Unadjust… got_cvd 6 ran_pars sd__(Intercept) 56.8 NA NA Unadjust… got_cvd 7 fixed (Intercept) 0.00000532 0.000000503 5.61e-5 Adjusted got_cvd 8 fixed total_cholesterol_scal… 1.21 0.347 4.22e+0 Adjusted got_cvd 9 fixed sexWoman 0.272 0.0143 5.15e+0 Adjusted got_cvd 10 ran_pars sd__(Intercept) 55.5 NA NA Adjusted got_cvd
Let's do the same thing with the list of adjusted models, but instead, we'll bind rows of both unadjusted and adjusted models. This puts all model items into a single dataframe. Lastly, let's add information about the outcome. We now have a single model dataframe that we can use to create plots and tables.
Let's practice wrangling!
Now you know how to wrangling model results. Let's practice!
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.