Medical information processing apparatus, medical information processing method, and recording medium

1 . A medical information processing apparatus comprising processing circuitry including a plurality of structure models corresponding respectively to a plurality of medical decisions, each structure model of the plurality of structure models being represented by an effect-followed posterior distribution and parameters defining the effect-followed posterior distribution,

wherein the processing circuitry is configured to:

calculate, for each of the plurality of medical decisions, an effect evaluation value using the corresponding effect-followed posterior distribution;

determine, from among the plurality of medical decisions, a medical decision to be assigned to a subject by executing a bandit algorithm that selects the medical decision having a highest effect evaluation value;

accumulate, in a database, observation data including an observation value of an effect produced for the subject by performing the determined medical decision;

update, in response to the accumulated observation data and as part of determining the medical decision to be assigned to the subject, the parameters of each of the plurality of structure models while retaining a structure of each of the plurality of structure models; and

update the structure of each of the plurality of structure models at a frequency lower than that for updating the parameters, the updating of the structure being performed in response to a determination that an amount of the accumulated observation data satisfies a predetermined structural update condition, the updating of the structure including at least one of:

division of each of the plurality of structure models into structure models corresponding to values which a feature amount of an attribute or a condition of the subject can take;

addition, deletion, or change of the feature amount for each of the plurality of structure models;

addition of a structure model for another medical decision; or

change of a type of a prior distribution representing each of the plurality of structure models.

2 . The medical information processing apparatus according to claim 1 , wherein the processing circuitry is further configured to update the structure at a timing where a reference number or more sets of the observation data pertaining to the feature amount are accumulated.

3 . The medical information processing apparatus according to claim 1 , wherein

the plurality of structure models include first structure models corresponding to values which a feature amount of the subject can take, and

the processing circuitry is further configured to, as the updating of the structure, combine the first structure models into a structure model not depending on the feature amount.

4 . The medical information processing apparatus according to claim 1 , wherein

the plurality of structure models include a first structure model in which a first feature amount of the subject is included, and

the processing circuitry is further configured to, as the updating of the structure, change the first structure model to a second structure model in which the first feature amount and a second feature amount are included.

5 . The medical information processing apparatus according to claim 4 , wherein the processing circuitry is further configured to update the structure at a timing where a reference number or more sets of the observation data pertaining to the second feature amount are accumulated.

6 . The medical information processing apparatus according to claim 1 , wherein

the plurality of structure models include a first structure model in which a first feature amount of the subject is included, and

the processing circuitry is further configured to, as the updating of the structure, change the first structure model to a second structure model in which a second feature amount is included in place of the first feature amount.

7 . The medical information processing apparatus according to claim 6 , wherein the processing circuitry is further configured to update the structure at a timing where a reference number or more sets of the observation data pertaining to the second feature amount are accumulated.

8 . The medical information processing apparatus according to claim 1 , wherein

the plurality of structure models include a first structure model in which a plurality of feature amounts are included, and

the processing circuitry is further configured to, as the updating of the structure, change the first structure model to a second structure model from which a predetermined portion of the feature amounts is deleted.

9 . The medical information processing apparatus according to claim 1 , wherein

the plurality of medical decisions include a first medical decision,

the plurality of structure models include a first structure model for handling the first medical decision, and

the processing circuitry is further configured to, as the updating of the structure, add a second structure model for handling a second medical decision different from the first medical decision, to the model.

10 . The medical information processing apparatus according to claim 9 , wherein the processing circuitry is further configured to add the structure model at a timing where a reference number or more sets of the observation data according to the second medical decision are accumulated.

11 . The medical information processing apparatus according to claim 1 , wherein

the plurality of medical decisions include a first medical decision and a second medical decision,

the model comprises a first structure model for handling the first medical decision and a second structure model for handling the second medical decision, and

the processing circuitry is further configured to, as the updating of the structure, discard the first structure model or the second structure model.

12 . The medical information processing apparatus according to claim 1 , wherein the processing circuitry is further configured to, as the updating of the structure, change a type of a prior distribution which the effect produced for the subject follows, the effect being attributable to a medical act corresponding to the medical decision.

13 . The medical information processing apparatus according to claim 1 , wherein the processing circuitry is further configured to update the structure at a timing where a reference number or more sets of the observation data are accumulated.

14 . The medical information processing apparatus according to claim 1 , wherein the processing circuitry is further configured to update the structure at a timing where a predetermined period of time elapses.

15 . The medical information processing apparatus according to claim 1 , wherein the processing circuitry is further configured to update the structure at a timing where the updating is expected to improve performance of the model.

16 . The medical information processing apparatus according to claim 1 , wherein the processing circuitry is further configured to update the parameter at a timing where a reference number or more sets of the observation data are accumulated or at a timing where a predetermined period of time elapses,

the parameter defining a posterior distribution of expected values of the effect produced for the subject, and

the effect being attributable to a medical act corresponding to the medical decision.

17 . A medical information processing method implemented by processing circuitry including a plurality of structure models corresponding respectively to a plurality of medical decisions, each structure model of the plurality of structure models being represented by an effect-followed posterior distribution and parameters defining the effect-followed posterior distribution, the method comprising:

calculating, for each of the plurality of medical decisions, an effect evaluation value using the corresponding effect-followed posterior distribution;

determining, from among the plurality of medical decisions, a medical decision to be assigned to a subject by executing a bandit algorithm that selects the medical decision having a highest effect evaluation value;

accumulating, in a database, observation data including an observation value of an effect produced for the subject by performing the determined medical decision;

updating, in response to the accumulated observation data and as part of determining the medical decision to be assigned to the subject, the parameters of each of the plurality of structure models while retaining a structure of each of the plurality of structure models; and

updating the structure of each of the plurality of structure models at a frequency lower than that for updating the parameters, the updating of the structure being performed in response to a determination that an amount of the accumulated observation data satisfies a predetermined structural update condition, the updating of the structure including at least one of:

division of each of the plurality of structure models into structure models corresponding to values which a feature amount of an attribute or a condition of the subject can take;

addition, deletion, or change of the feature amount for each of the plurality of structure models;

addition of a structure model for another medical decision; or

change of a type of a prior distribution representing each of the plurality of structure models.

18 . A non-transitory computer-readable recording medium storing instructions executable by processing circuitry including a plurality of structure models corresponding respectively to a plurality of medical decisions, each structure model of the plurality of structure models being represented by an effect-followed posterior distribution and parameters defining the effect-followed posterior distribution, the instructions, when executed by the processing circuitry, cause the device to:

calculate, for each of the plurality of medical decisions, an effect evaluation value using the corresponding effect-followed posterior distribution;

determine, from among the plurality of medical decisions, a medical decision to be assigned to a subject by executing a bandit algorithm that selects the medical decision having a highest effect evaluation value;

accumulate, in a database, observation data including an observation value of an effect produced for the subject by performing the determined medical decision;

update, in response to the accumulated observation data and as part of determining the medical decision to be assigned to the subject, the parameters of each of the plurality of structure models while retaining a structure of each of the plurality of structure models; and

update the structure of each of the plurality of structure models at a frequency lower than that for updating the parameters, the updating of the structure being performed in response to a determination that an amount of the accumulated observation data satisfies a predetermined structural update condition, the updating of the structure including at least one of:

division of each of the plurality of structure models into structure models corresponding to values which a feature amount of an attribute or a condition of the subject can take;

addition, deletion, or change of the feature amount for each of the plurality of structure models;

addition of a structure model for another medical decision; or

change of a type of a prior distribution representing each of the plurality of structure models.

19 . A medical information processing apparatus comprising

processing circuitry configured to:

calculate, using a model capable of calculating an effect evaluation value corresponding to a medical decision and represented by an effect-followed posterior distribution and parameters defining the effect-followed posterior distribution, the effect evaluation value corresponding to the medial decision;

determine the medical decision to be assigned to a subject by executing a bandit algorithm that selects the medical decision having a highest effect evaluation value;

acquire an effect observation value for an effect produced for the subject by applying a medical act corresponding to the determined medical decision;

accumulate, in a database, observation data including an identifier for the subject, the determined medical decision, and the acquired effect observation value; and

update the model based on the accumulated observation data,

wherein the processing circuitry comprises:

a first updater configured, in response to the accumulated observation data and as part of determining the medical decision to be assigned to the subject, to update the parameters of the model while retaining a structure of the model; and

a second updater configured, at a frequency lower than a frequency of updating the parameters and in response to a determination that an amount of the accumulated observation data satisfies a predetermined structural update condition, to update the structure of the model by at least one of:

division of the model into a plurality of models corresponding to values which a feature amount of an attribute or a condition of the subject can take;

addition, deletion, or change of a feature amount included in the model;

addition of a model corresponding to another medical decision; or

change of a type of a prior distribution representing the model.