-
Notifications
You must be signed in to change notification settings - Fork 647
Generic Module Framework: Transitions
The generic module framework currently supports the following transitions:
- Direct
- Distributed
- Conditional
- Complex
- Table
- TypeOfCare Experimental
Direct
transitions are the simplest of transitions. They transition directly to the indicated state. The value of a direct_transition
is simply the name of the state to transition to.
Please note that, for simplicity, transition examples in this document will always start at an Initial
state. In real modules, transitions can be applied to any state (except a Terminal
state) and can transition to any state.
The following example demonstrates a state that should transition directly to the "Delay_For_Encounter"
state:
"Initial": {
"type": "Initial",
"direct_transition": "Delay_For_Encounter"
}
Distributed
transitions will transition to one of several possible states based on the configured distribution. Distribution values are from 0.0
to 1.0
, such that a value of 0.55
would indicate a 55% chance of transitioning to the corresponding state. A distributed_transition
consists of an array of distribution
/transition
pairs for which the distribution
values should sum up to 1.0
.
If the distribution
values do not sum up to 1.0, the remaining distribution will transition to the last defined transition
state. For example, given distributions of 0.3
and 0.6
, the effective distribution of the last transition will actually be 0.7
.
If the distribution
values sum up to more than 1.0
, the remaining distributions are ignored (for example, if distribution values are 0.75
, 0.5
, and 0.3
, then the second transition will have an effective distribution of 0.25
, and the last transition will have an effective distribution of 0.0
).
The following example demonstrates a state that should transition to the "Medication_1"
state 15% of the time, the "Medication_2"
state 55% of the time, and the "Medication_3"
state 30% of the time:
"Initial": {
"type": "Initial",
"distributed_transition": [
{
"distribution": 0.15,
"transition": "Medication_1"
},
{
"distribution": 0.55,
"transition": "Medication_2"
},
{
"distribution": 0.30,
"transition": "Medication_3"
}
]
}
For cases where transition probabilities are likely to change based on many different factors, it may be useful to use a "named" transition probability, where the probability of taking any transition is based on the value in an attribute rather than fixed. In this case, instead of a number, the distribution
on the transition option is an object containing the name of the attribute to look up the transition probability, and a default value for the case where the attribute is not present on the patient.
"distributed_transition": [
{
"distribution": { "attribute" : "probability1", "default" : 0.15 },
"transition": "Terminal1"
},
{
"distribution": { "attribute" : "probability2", "default" : 0.55 },
"transition": "Terminal2"
},
{
"distribution": { "attribute" : "probability3", "default" : 0.30 },
"transition": "Terminal3"
}
]
Conditional
transitions will transition to one of several possible states based on conditional logic. A conditional_transition
consists of an array of condition
/transition
pairs which are tested in the order they are defined. The first condition that evaluates to true
will result in a transition to its corresponding transition
state. The last element in the condition_transition
array may contain only a transition
(with no condition
) to indicate a "fallback transition" when all other conditions are false
.
If none of the conditions
evaluated to true
, and no fallback transition was specified, the module will transition to a default Terminal
state.
Please see the Logic section for more information about creating logical conditions.
The following example demonstrates a state that should transition to the "Male_Patient"
state for male patients, the "Female_Patient
state for female patients, and the "Unknown_Gender"
state for patients with an unrecognized gender value.
"Initial": {
"type": "Initial",
"conditional_transition": [
{
"condition": {
"condition_type": "Gender",
"gender": "M"
},
"transition": "Male_Patient"
},
{
"condition": {
"condition_type": "Gender",
"gender": "F"
},
"transition": "Female_Patient"
},
{
"transition": "Unknown_Gender"
}
]
}
Complex
transitions are a combination of direct, distributed, and conditional transitions. A complex_transition
consists of an array of condition/transition pairs which are tested in the order they are defined. The first condition that evaluates to true
will result in a transition based on its corresponding transition
or distributions
. If the module defines a transition
, it will transition directly to that named state. If the module defines distributions
, it will then transition to one of these according to the same rules as the distributed_transition
. See Distributed for more detail. The last element in the complex_transition
array may omit the condition
to indicate a fallback transition when all other conditions are false
.
If none of the conditions
evaluated to true
, and no fallback transition was specified, the module will transition to the last defined transition.
Please see the Logic section for more information about creating logical conditions.
Example
The following example demonstrates a state that for male patients should transition to the "Male_Medication_1"
state with 15% probability and the "Male_Medication_2"
state with 85% probability, and for female patients should transition directly to the "Female_Medication"
state.
"Initial": {
"type": "Initial",
"complex_transition": [
{
"condition": {
"condition_type": "Gender",
"gender": "M"
},
"distributions": [
{
"distribution": 0.15,
"transition": "Male_Medication_1"
},
{
"distribution": 0.85,
"transition": "Male_Medication_2"
}
]
},
{
"condition": {
"condition_type": "Gender",
"gender": "F"
},
"transition": "Female_Medication"
}
]
}
Table-based transitions are used for probabilities that vary widely between different segments or cohorts of the population. When the probability of an event occurring is based on any combination of patient attributes (e.g. race
, ethnicity
, gender
, age
, smoker
status, or any other attribute in the system) you can use a lookup_table_transition
.
A lookup_table_transition
is defined as a JSON array that contains one or more transitions, each with a transition
, a default_probability
, and the lookup_table_name
to use to lookup transition probabilities.
The transition
is the name of the next State to transition into, and the default_probability
is a number (0.0 - 1.0
) that represents the default probability of using this transition in the case that the lookup table contains no match.
The lookup table itself should have at least N + M
columns, where M
is the number of transitions. The last M
columns must have headers that match the transition names (i.e. declared in the JSON) and the values in those columns must be numeric probabilities between 0.0 - 1.0
. The first N
columns can be any subset of patient attributes.
This example consists of both JSON which defines the table-based transition in the module, and a comma-separated value (CSV) file that defines the lookup table.
"Determine_Condition": {
"type": "Simple",
"name": "Determine_Condition",
"lookup_table_transition": [
{
"transition": "Lookuptablitis",
"default_probability": "0",
"lookup_table_name": "lookuptablitis.csv"
},
{
"transition": "No_Lookuptablitis",
"default_probability": "1",
"lookup_table_name": "lookuptablitis.csv"
}
]
}
The contents of lookuptablitis.csv
are below:
age,gender,state,Lookuptablitis,No_Lookuptablitis
0-17,M,Massachusetts,0,1
18-44,M,Massachusetts,0.25,0.75
45-64,M,Massachusetts,0.5,0.5
64-140,M,Massachusetts,0.75,0.25
0-17,F,Massachusetts,0,1
18-44,F,Massachusetts,0.3,0.7
45-64,F,Massachusetts,0.8,0.2
64-140,F,Massachusetts,0,1
...
Note - this is an experimental feature! It is only available on the virtual-medicine-transition branch
TypeOfCare
transitions are intended to be used to allow different care paths depending on what type of care a patient is likely to seek. The transition has three different possible transitions: ambulatory
, emergency
and telemedicine
.
The transition selected will be a random weighted selection driven by properties in the src\main\resources\telemedicine_config.json
file. Broadly, transition likelihood is driven by the person's insurance (or lack thereof) and the time in the simulation. The configuration file allows users to specify a year during which telemedicine encounters should start. By default, the start_year
property is set to 2020
to reflect the rapid increase in the use of telemedicine during the COVID-19 pandemic. It is expected that each of these encounters would offer different care pathways based on their mode. For example, the telemedicine pathway would not include any lab observations.
The following example demonstrates a state that will randomly transition to a PCP Encounter, Telemedicine Encounter or ED Visit.
"Initial": {
"type": "Initial",
"type_of_care_transition": {
"ambulatory": "PCP_Encounter",
"telemedicine": "Virtual_Encounter",
"emergency": "ED_Encounter"
}
}