## Authors

- David Topps
- Corey Wirun
- Michelle Cullen
- Rachel Ellaway

This is part of a project looking at the factors that clinicians might take into consideration when making clinical decisions. This is a complex and multifactorial process, which is sometimes performed in a step-wise algorithmic manner, but more commonly as a less well-defined heuristic process.

In evaluating each therapeutic action or step, we will need to consider various factors as to whether the step was useful or not. Ideally, these factors should all be present and affected by each step in the case but this may not always be true.

Can we create a matrix of factors or variables that will help with scenario design? This page is based on a new technique where we initially used a collaboratively-edited Google Doc so that our team could edit and contribute to the suggested factors, discuss them via edits and comments.

In Structural Equation Modeling, we have latent variables, observed variables, independent variables and dependent variables, just as with any other regression model. Can we also deal with expected changes in those variables? E.g. if we add HCTZ for BP, how much do we expect the BP to drop?

In this mini-study, we look at a number of common clinical decisions and surveyed our clinician team to consider what factors they might take into account.

##### BP meds

- Cost
- Expected BP drop
- Probability of side fx
- Dose frequency
- Compliance probability (might be latent resulting from cost, side effects and dose frequency)
- Familiarity with name or use (also likely latent)

##### Lipid meds

- Cost
- Expected lipid drop
- Probability of side fx
- Dose frequency
- Compliance probability (might be latent resulting from cost, side effects and dose frequency)
- Familiarity with name or use (also likely latent)

##### Vertigo measures

- Effectiveness
- How easy to administer or order
- How easy for patient to do
- Cost to patient
- Cost to AHS
- How many times?
- Compliance probability (might be latent resulting from cost, side effects and dose frequency)
- Familiarity with name or use (also likely latent)

##### Heartburn measures

- Effectiveness
- How easy to administer or order
- How easy for patient to do
- Probability of side fx
- Cost to patient
- Cost to AHS
- How many times?
- Compliance probability (might be latent resulting from cost, side effects and dose frequency)
- Familiarity with name or use (also likely latent)

##### Weight loss

- Effectiveness
- How easy to administer or order
- How easy for patient to do
- Probability of side fx
- Cost to patient
- Cost to AHS
- How many times?
- Compliance probability (might be latent resulting from cost, side effects and dose frequency)
- Familiarity with name or use (also likely latent)

##### Wearing a helmet

- Effectiveness
- How easy to administer or order
- How easy for patient to do
- Cost to patient
- Cost to AHS
- How many times?
- Compliance probability (might be latent resulting from cost, side effects and dose frequency)
- Familiarity with name or use (also likely latent)

##### Smoking cessation measures

- Effectiveness
- How easy to administer or order
- How easy for patient to do
- Cost to patient
- Cost to AHS
- How many times?
- Compliance probability (might be latent resulting from cost, side effects and dose frequency)
- Familiarity with name or use (also likely latent)

##### Back pain measures

- Effectiveness
- How easy to administer or order
- Including the risk that pharm or someone else contradicts this

- How easy for patient to do
- Cost to patient
- Cost to AHS
- How many times?
- Compliance probability (might be latent resulting from cost, side effects and dose frequency)
- Familiarity with name or use (also likely latent)
- Interfering with other activities e.g. sex, alcohol
- Stigma attached to measure or drug
- Risk of off-label use
- Probability of side fx

##### Choosing an Abx

- Effectiveness
- How easy to administer or order
- Including the risk that pharm or someone else contradicts this

- How easy for patient to do
- Cost to patient
- Cost to AHS
- How many times?
- Compliance probability (might be latent resulting from cost, side effects and dose frequency)
- Familiarity with name or use (also likely latent)
- Interfering with other activities e.g. sex, alcohol
- Stigma attached to measure or drug
- Risk of off-label use
- Probability of side fx

##### Anticoagulant measures

- Monitoring costs to AHS
- Monitoring costs/convenience to patient
- Although that could be grouped under ‘how easy to do’
- Or combined as part of compliance probability

- Probability of side fx
- Effectiveness
- How easy to administer or order e.g. triplicate Rx
- Including the risk that pharm or someone else contradicts this

- How easy for patient to use
- Cost to patient
- Cost to AHS
- How many times?
- Compliance probability (might be latent resulting from cost, side effects and dose frequency)
- Familiarity with name or use (also likely latent)
- Interfering with other activities e.g. sex, alcohol, salads

##### Choosing an analgesic

- Effectiveness
- How easy to administer or order e.g. triplicate Rx
- Including the risk that pharm or someone else contradicts this

- How easy for patient to use
- Cost to patient
- Cost to AHS
- How many times?
- Compliance probability (might be latent resulting from cost, side effects and dose frequency)
- Familiarity with name or use (also likely latent)
- Interfering with other activities e.g. sex, alcohol
- Stigma attached to measure or drug
- Risk of off-label use
- Probability of side fx

##### Choosing an antidepressant or behavioural health measures

- Effectiveness
- How easy to administer or order
- Including the risk that pharm or someone else contradicts this

- How easy for patient to use
- Cost to patient
- Cost to AHS
- How many times?
- Compliance probability (might be latent resulting from cost, side effects and dose frequency)
- Familiarity with name or use (also likely latent)
- Interfering with other activities e.g. sex, alcohol
- Stigma attached to measure or drug
- Risk of off-label use
- Probability of side fx

## Lumping and Splitting

One of the issues that we ran into when trying to get agreement amongst our expert clinician group was the classic problem of ‘lumping and splitting’. Some experts preferred to split some of these factors into sub-sets; some preferred to lump factors into groups. This is a problem with any ontology: https://en.wikipedia.org/wiki/Lumpers_and_splitters – the process is generally part of ontology normalization.

Lord and Stevens, in their paper ‘Facets, Tiers and Gems: Ontology Patterns for Hypernormalisation’,

http://ceur-ws.org/Vol-2137/paper_21.pdf , provide a very helpful overview of this problem. It describes various approaches, including facets, gems, tiers, and the downsides of older rigid tree-style hierarchies.

One possible solution to consider is a faceted ontology (aka folksonomy or tagging). Lord & Stevens address this in their paper, as well as more sophisticated things like gems (groups of facets – a neat term).

As we examine how clinicians weigh various factors in a clinical decision-making process, we will inevitably run into small disputes about how to group and classify these factors. Facets, gems, tiers and annotations may all play a part in this.

We created a small graph database which illustrates this challenge.