Injury Event Calculation

Injuries (fatal, serious, and slight) are modelled as a non-linear function of changes in distance by 'victim mode' and 'striking vehicle'.

Calculation https://github.com/ITHIM/ITHIM-R/issues/9

Stratification

1. age group (striker/ victim) https://github.com/ITHIM/ITHIM-R/issues/8
2. gender (striker/ victim)
3. mode (striker/ victim)
4. road type
5. injury severity Events are stratified by road type (motorway, major, minor) if the exposure and event data are available. Changes in distance are (separately for strikers and victims) summed over age group and gender but not road type. Road type is most important for regional models, as replacing short trips by walking or cycling will replace distance on local roads rather than motorways. For an urban model this is probably less of a problem.

A major challenge is going between person travel (used for PA and exposure as victims) and vehicle travel https://github.com/ITHIM/ITHIM-R/issues/15

---

How the module will work:

---

Data inputs

There will be two or three inputs.

1. The first input comes directly from the synthetic population. It will be a data frame with known headers, and it will detail all the travel done by each demographic group by each mode. (tbi when we have agreed (a) what the synthetic population looks like and (b) what the processed version looks like. The processed version might be used also by pollution and noise modules.) There are multiple versions of this input: one per scenario (where the baseline counts as one scenario). This object is described here in the overview.
2. The second input is an injury data set. This will likely be a spreadsheet or Rdata (tbi) object with prespecified column and row names. Row names will be ICD-10 codes. Column headers will be multivariate and will specify the covariates.
3. (tbi) The optional data input is a set of additional data that aren't captured in the synthetic population, e.g. bus and HGV driving. Perhaps also other covariates like road type can be covered here. This might also be included in the object described here.

Optional user inputs

There will be optional user inputs, such as which model type to use, which age groups to use if the model is large (tbi). The user can also upload previously saved models.

Functionality

The first step (tbi) will be to check that the inputs make sense, and to bounce back to the user if they don't. The next step (tbi) is data harmonisation, e.g. if there are different age groupings in the distance and injury datasets.

Two models run in parallel: (a) one for collisions, where we have data relating to striking vehicles, and (b) one for no/other/unknown vehicle, for any injuries that are caused by things without drivers we could model.

If the datasets are small, the module tries all possible models for each of (a) and (b) and chooses the best in terms of AIC. If datasets are larger, we try a subset of possible models (tbi).

The 'predict' function generates predicted injury numbers for each input of type 1, i.e. one per scenario.

Output

The output is a set of arrays (one per scenario) of all injuries caused, according to the covariates supplied in the inputs. These will be handed to a GBD calculator (tbi), described here in the overview.

The first array corresponds to the baseline scenario, and corresponds to a multidimensional set of who-hit-whom matrices, which is described here in the overview.

Users can save models as they go along.

---

To-do list

1. Find a few example datasets
2. Generalise engine to accommodate any conceivable dataset combination
3. Fix the form of the processed synthetic population
4. Choose way to build model for larger datasets
5. What to do if data are fatality only
6. Accommodate additional distance dataset
7. What to do if there are e.g. bus distance data but not HGV
8. V00 code
9. Add parameter for underreporting
10. Think about safety in numbers