11 Oracle Healthcare Data Model Data Mining Models

This chapter provides reference information about the data mining model provided with Oracle Healthcare Data Model.

This chapter includes the following sections:

• About Data Mining in Oracle Healthcare Data Model

• Model 1: Weight of Evidence Over Adverse Event (Patient Fall)

About Data Mining in Oracle Healthcare Data Model

Oracle Healthcare Data Model mining model includes data mining packages, source tables (MV) and target tables. The source tables are extracted from Oracle Healthcare Data Model schema and are used to train the models. The target tables contain the mining result data, for example, mined rules. Data mining packages pull in the source data, feed it into the data mining packages, and populate the target tables with the results. The data in the target tables can be presented in reports.

Note:

Oracle does not support modified or new data models. Consequently, do not change the data models that are defined and delivered with Oracle Healthcare Data Model. If changes are required to meet your organization requirements, create a copy of the delivered Oracle Healthcare Data Model where you can make your changes.

As shown in Table 11-1, the Oracle Healthcare Data Model mining model uses the specified algorithms for the specific problem.

Table 11-1 Oracle Healthcare Data Model Algorithm Types Used by Model

Model	Algorithms Used by Data Mining Model
Model 1: Weight of Evidence Over Adverse Event (Patient Fall)	Predictive Model Support Vector Machine (SVM)

Understanding the Mining Model Schema and Architecture

The Oracle Healthcare Data Model mining consists of one schema: ohdm_sys. All the objects from Oracle Healthcare Data Model Mining models are included in the Oracle Healthcare Data Model schema (ohdm_sys). The Oracle Healthcare Data Model mining objects include:

• Mining Model Package (pkg_ohdm_mining): Given source data in the mining source table, the mining package generates Mined Rules, Predicted Results, and additional information.

• Mining Model Source Table: presents useful Oracle Healthcare Data Model information to Oracle Mining algorithms as one table. For more information, see Table 11-2.

• Mining Result Tables (synonyms): Mining Result Tables save the output from Mining models, which is normally produced from the mining apply process.

• Mining Model Support Tables: The mining model support tables are primarily intermediate tables used during the mining model creation or testing process. Most of the mining model support tables have names that start with "DM$".

Note:

Do not delete the mining model support tables; the DM$ tables can be very difficult to reconstruct if they are deleted.

The ohdm_sys schema includes the following:

• Oracle Healthcare Data Model Main Model, which are the Oracle Healthcare Data Model. For more information, see Chapter 4, "Oracle Healthcare Data Model Physical Data Model".

• Mining Source Tables, Mining Result Tables, and Mining Support tables. These tables are also created in ohdm_sys schema.

• Mining Model Package.

Using the Mining Model Refresh Procedure

Over time, the Oracle Healthcare Data Model database information and behavior may change. Therefore, you may want to refresh the trained mining models based on the latest stored usage data. For more information about the Oracle Mining training and Scoring (applying) process, see Oracle Data Mining Concepts.

To refresh the mining model based on latest data, call the procedure pkg_ohdm_mining.refresh_mining_source. This procedure performs the following tasks:

1. Refreshes the mining source table dwd_advr_evt_fall.

2. Creates a predictive model: advr_fall_mod and also deciphers the fall factors for each patient.

Model Creation Trouble Shooting

The errors that occur during mining model refresh are saved into the table named: DWC_INTRA_ETL_ACTVTY as is other standard Oracle Healthcare Data Model Intra-ETL package errors and information.

If you see privilege errors check the following privileges are granted to the OHDM_SYS user:

GRANT all on dbms_data_mining to ohdm_sys;
GRANT CREATE MINING MODEL TO ohdm_sys;
GRANT CREATE JOB TO ohdm_sys;
GRANT CREATE PROCEDURE TO ohdm_sys;
GRANT CREATE SEQUENCE TO ohdm_sys;
GRANT CREATE SESSION TO ohdm_sys;
GRANT CREATE SYNONYM TO ohdm_sys;
GRANT CREATE TABLE TO ohdm_sys;
GRANT CREATE TYPE TO ohdm_sys;
GRANT CREATE VIEW TO ohdm_sys;
GRANT EXECUTE ON ctxsys.ctx_ddl TO ohdm_sys;

Model 1: Weight of Evidence Over Adverse Event (Patient Fall)

The Weight of Evidence over Adverse Event (Patient Fall) mining model does the following:

• Gathers all information about patient falls from ohdm_sys schema and puts the information together in a single table.

• Runs the Oracle Mining Prediction algorithm to create a predictive model. This model tries to predict future fall probability for a new patient according to the known information about the patient. The prediction is based on past patient falls.

• Deciphers the risk factors for each patient.

• Creates a report that provides the contributing fall risk factors for a patient and identifies the weight associated with each factor.

• The model also runs the prediction for recently admitted patients to identify the patients with a higher fall probability, so that the service provider can take preventive actions.

Patient Fall: Mining Source Table Attributes

Table 11-2 shows the attributes identified from Oracle Healthcare Data Model as input source variables for the mining model. The major source data for this mining model is from the source table, dwd_advr_evt_fall.

Table 11-2 dwd_advr_evt_fall Mining Source Table Attributes

Attribute	Description
OBSV_ID	ID for the Observation
ENC_ID	ID for the ENCOUNTER
PT_ID	ID for the Patient
PT_AGE	Patient AGE
FALL_CNT_ENC	History of Falls in encounter
FALL_CNT_3M	History of Falls Within 3 months (during hospital encounter)
FALL_CNT_12M	History of Falls Within 3-12 months
FALL_CNT_ALL	History of Falls Within more than 12 months
SUBABS_CNT_ENC	Substance Abuse history of Falls in encounter
SUBABS_CNT_3M	Substance Abuse history Within 3 months (during hospital encounter)
SUBABS_CNT_12M	Substance Abuse history Within 3-12 months
SUBABS_CNT_ALL	Substance Abuse history Within more than 12 months
DIAG_ANXIETY	Diagnosis-- Anxiety
DIAG_DEPRESSION	Diagnosis-- Depression
DIAG_PARKINSON	Diagnosis-- Parkinson's
DIAG_ABNORM_GAIT	Diagnosis-- Abnormality of Gait
DIAG_STROKE	Diagnosis-- Stroke
BARTHEL_SCALE	BARTHEL SCALE
AMBULATORY_AID	Ambulatory Aid Observation, like Bedrest, wheel chair, Nurse assist, …
IV	IV
IVR	Heparin Lock
GAIT	Gait, like Normal, bedrest, immobile
MENTAL_STATUS	Mental Status, like Oriented, Mildly Impaired, …
MEDIC_AMT_SEDATIVE	Medication -- Sedative
MEDIC_AMT_ANTI_DEPRESSANT	Medication -- Anti-Depressants
MEDIC_AMT_ANTI_PARKINSON	Medication -- Anti-Parkinson s
MEDIC_AMT_DIURETICS	Medication -- Diuretics
MEDIC_AMT_ANTI_HYPERTENSIVE	Medication -- Anti-hypertensives
MEDIC_AMT_HYPNOTICS	Medication -- Hypnotics
MEDIC_AMT_ANTIARRHYTHMICS	Medication -- Antiarrhythmics
MEDIC_AMT_OPIATE	Medication -- Opiate
PSYCH_ANXIETY	Psychological Observation -- Anxiety
PSYCH_DEPRESSION	Psychological Observation -- Depression
PSYCH_DEC_COOPERATION	Psychological Observation -- Decrease in Cooperation
PSYCH_DEC_INSIGHT	Psychological Observation -- Decrease in Insight
PSYCH_DEC_JUDGMENT	Psychological Observation -- Decrease in Judgment
PSYCH_DEC_ADL	Psychological Observation -- Decrease in performing ADLs
PSYCH_FALLS	Psychological Observation -- Falls
PSYCH_AGITATED	Psychological Observation -- Agitated
VISUAL_IMPAIR	Visual Impairment
HEAR_IMPAIR	Hearing impairment
URINE_ASSESS	Urinary Assessment
FACILITY	ID for facilities, locations, and so on.

Patient Fall Factors: the Ranking Output

The mined results are saved into the target table. The weight of each fall factor is saved into the target table: dwd_advr_woe after each successful model refresh. Table 11-3 shows the target table structure.

Table 11-3 Patient Fall Output Table dwd_advr_woe Columns

Column Name	Description
case_id	The case ID inherited from mining source table.
attribute_name	Name of the fall factor
Weight	Weight of the fall factor
Rank	Rank of the fall factor, reserved for future use.

Patient Fall: Prediction Algorithm

For more information on mining algorithms, see Oracle Data Mining Concepts and Oracle Data Mining Application Developer's Guide.