Chapter 21: Using Exploratory Visualization In The Analysis Of Medical Product Safety In Observational Healthcare Data by Patrick Ryan

Guidance for the creation of figures using Spotfire.

Figures 21.1-21.4: Patient Profile

Data: PEDRO_summary

The person profile visualizations are created from a standardize dataset with one row per observation date. Note, one observation (ex. period of drug exposure) will have two records: drug start date and drug end date. The columns in the data table are:

· PERSON_ID – unique identifier for each individual in the dataset

· TABLE_NAME – string that contains name of OMOP common data model table (http://omop.fnih.org/CDMvocabV4) from which the observation was extracted

· DAYS_TO_INDEX – each person qualifies for a cohort on a given index date (ex: persons with a diagnosis have an index date at the date of first diagnosis). DAYS_TO_INDEX records the distance between the observation date to the index date. Negative numbers occur prior to index, while positive number occur after index, and index date = 0.

· CONCEPT_ID – unique identifier for the type of observation, using the OMOP vocabulary (http://omop.fnih.org/CDMvocabV4)

· CONCEPT_NAME – the string corresponding to the unique concept identifier, from the lookup in the CONCEPT table of the OMOP vocabulary

· CONCEPT_TYPE – classification of observations, based on relation to target exposure and outcome. In addition to specifying the ‘target exposure’ and ‘target outcome’, the OMOP vocabulary can be used to identify ‘Related’ concepts, such as ‘Target exposure FDA-approved indication’, ‘Target exposure off-label use’, and ‘Target exposure contraindication’, ‘Target exposure indication FDA-approved alternative treatment’, ‘Target exposure indication off-label alternative treatment’, ‘Target outcome FDA-approved indicated treatment’, ‘Target outcome off-label use’, ‘Target outcome contraindicated treatment’, and ‘Target outcome sibling concept’.

· CONCEPT_TYPE_ORDER – numeric value to ensure ‘Target outcome’ and ‘Target exposure’ are larger sized icons than ‘Related’ or ‘Other’ concepts.

· TIME_TO_EVENT – the duration from the index date (cohort entry) until the event date.

· DATE_TYPE – specify whether observation date is ‘START DATE’ or ‘END DATE’.

Visualization type: Scatter Plot

Properties:

· Data: PEDRO_summary

· Appearance: Jittering on y-axis, None on x-axis. ‘Use the same amount of jittering for both X and Y’ unchecked

· X-axis: DAYS_TO_INDEX

· Y-axis: TABLE_NAME

· Colors: CONCEPT_TYPE

· Size: CONCEPT_TYPE_ORDER

· Shape: CONCEPT_TYPE, DATE_TYPE. If CONCEPT_TYPE = ‘Target outcome’, then shape is Star, else use right arrow for ‘Start date’ and left arrow for ‘End date’

· Labels: CONCEPT_NAME

· Tooltip: PERSON_ID, TABLE_NAME, CONCEPT_NAME, DATE_TYPE, DAYS_TO_INDEX, CONCEPT_TYPE

· Legend: ‘Show legend’ checked. Display ‘Color by’

· Trellis: Rows and columns/Pages: PERSON_ID

· Line Connection: None

· Market By: Row Number

· Lines & Curves: None

· Error Bars: None

Figure 21.5: Cohort Profile

Visualization type: Scatter Plot

Properties:

· Data: PEDRO_summary

· Appearance: None

· X-axis: DAYS_TO_INDEX

· Y-axis: TIME_TO_EVENT, PERSON_ID

· Colors: CONCEPT_TYPE

· Size: CONCEPT_TYPE_ORDER

· Shape: CONCEPT_TYPE, DATE_TYPE. If CONCEPT_TYPE = ‘Target outcome’, then shape is Star, else use right arrow for ‘Start date’ and left arrow for ‘End date’

· Labels: CONCEPT_NAME

· Tooltip: PERSON_ID, TABLE_NAME, CONCEPT_NAME, DATE_TYPE, DAYS_TO_INDEX, CONCEPT_TYPE

· Legend: ‘Show legend’ checked. Display ‘Color by’, ‘Shape by’

· Trellis: None

· Line Connection: None

· Market By: Row Number

· Lines & Curves: None

· Error Bars: None

Figure 21.6: Prevalence Treemap

Data: OSCAR_drug_treemap

The dataset is derived from summary statistics generated by OSCAR (http://omop.fnih.org/oscar). OSCAR is executed for each database and the summary statistics are compiled in a central coordinating center. Each row in the table represents a database – drug pair.

· SOURCE_ID: Unique identifier for the source database

· SOURCE_NAME: String description of the source database

· DRUG_CONCEPT_ID: unique identifier for the drug, using the OMOP vocabulary (http://omop.fnih.org/CDMvocabV4)

· DRUG_CONCEPT_NAME: the string corresponding to the unique concept identifier, from the lookup in the CONCEPT table of the OMOP vocabulary

· STANDARD_PREV: standardized prevalence estimates, derived by taking age/gender stratified prevalence estimates and standardized to the general population.

· RR: relative risk of the standardized prevalence, relative to the average standardized prevalence of the entire data network

· ATC7_CONCEPT_NAME, ATC5_CONCEPT_NAME, ATC3_CONCEPT_NAME, ATC1_CONCEPT_NAME: levels of the ATC drug classification system, as derived from mapping the DRUG_CONCEPT_ID into the OMOP vocabulary

Visualization: Treemap

· Data: OSCAR_drug_treemap

· Colors: Columns: Sum(RR); Color mode: Segments; One scale per: (None)

· Size: Sum(STANDARD_PREV)

· Hierarchy: ATC1_CONCEPT_NAME, ATC3_CONCEPT_NAME, ATC5_CONCEPT_NAME, ATC7_CONCEPT_NAME

· Labels: Show hierarchy headers, Show labels

· Tooltip: ATC7_CONCEPT_NAME, STANDARD_PREV, RR

· Legend: ‘Show legend’ checked, Display ‘Color by’

· Trellis: Rows and columns/Pages: SOURCE_NAME

Figure 21.7: Prevalence Treemap

Data: OSCAR_condition_treemap

The dataset is derived from summary statistics generated by OSCAR (http://omop.fnih.org/oscar). OSCAR is executed for each database and the summary statistics are compiled in a central coordinating center. Each row in the table represents a database – condition pair.

· SOURCE_ID: Unique identifier for the source database

· SOURCE_NAME: String description of the source database

· CONDITION_CONCEPT_ID: unique identifier for the condition, using the OMOP vocabulary (http://omop.fnih.org/CDMvocabV4)

· CONDITION_CONCEPT_NAME: the string corresponding to the unique concept identifier, from the lookup in the CONCEPT table of the OMOP vocabulary

· STANDARD_PREV: standardized prevalence estimates, derived by taking age/gender stratified prevalence estimates and standardized to the general population.

· RR: relative risk of the standardized prevalence, relative to the average standardized prevalence of the entire data network

· PT_CONCEPT_NAME, HLT_CONCEPT_NAME, HLGT_CONCEPT_NAME, SOC_CONCEPT_NAME: levels of the MedDRA classification system, as derived from mapping the CONDITION_CONCEPT_ID into the OMOP vocabulary

Visualization: Treemap

· Data: OSCAR_condition_treemap

· Colors: Columns: Sum(RR); Color mode: Segments; One scale per: (None)

· Size: Sum(STANDARD_PREV)

· Hierarchy: SOC_CONCEPT_NAME, CONDITION_CONCEPT_NAME

· Labels: Show hierarchy headers, Show labels

· Tooltip: CONDITION_CONCEPT_NAME, STANDARD_PREV, RR

· Legend: ‘Show legend’ checked, Display ‘Color by’

· Trellis: Rows and columns/Pages: SOURCE_NAME

Figure 21.8: Subgroup Trellis Scatterplot

Data: OSCAR_stratified_drug_prev

The dataset is derived from summary statistics generated by OSCAR (http://omop.fnih.org/oscar). OSCAR is executed for each database and the summary statistics are compiled in a central coordinating center. Each row in the table represents a prevalence of a drug in a database, with a given year/age/gender strata.

· SOURCE_ID: Unique identifier for the source database

· SOURCE_NAME: String description of the source database

· DRUG_CONCEPT_ID: unique identifier for the drug, using the OMOP vocabulary (http://omop.fnih.org/CDMvocabV4)

· DRUG_CONCEPT_NAME: the string corresponding to the unique concept identifier, from the lookup in the CONCEPT table of the OMOP vocabulary

· YEAR, AGE_GROUP, GENDER: stratification variables for which the prevalence is calculated

· PERSON_COUNT: number of persons in the database within the year/age/gender strata with exposure to the drug

· PERSON_PCT: percentage of persons in the database within the year/age/gender strata who have exposure to the drug

Visualization: Scatterplot

· Data: OSCAR_stratified_drug_prev

· Appearance: None

· X-axis: AGE_GROUP

· Y-axis: PERSON_PCT

· Colors: GENDER

· Size: None

· Shape: GENDER

· Labels: None

· Tooltip: SOURCE_NAME, DRUG_CONCEPT_NAME, YEAR, AGE_GROUP, GENDER, PERSON_COUNT, PERSON_PCT

· Legend: ‘Show legend’ checked. Display ‘Color by’

· Trellis: Rows and columns/Rows: SOURCE_NAME; Columns: DRUG_CONCEPT_NAME, YEAR

· Line Connection: None

· Market By: Row Number

· Lines & Curves: None

· Error Bars: None

Figure 21.9: Trellis Forest Plot

Data: HOI_score_full

The dataset is derived from OMOP experimental results, available at: (http://omop.fnih.org/2010ExperimentResults). Each method was executed within each database for each drug-outcome pair to generate effect estimates with associated standard error. Each row in the table represents one estimate from one method/analysis for one drug-outcome pair within one database.

· SOURCE_ID: Unique identifier for the source database

· SOURCE_NAME: String description of the source database

· DOI_CONCEPT_ID: unique identifier for the drug, using the OMOP vocabulary (http://omop.fnih.org/CDMvocabV4)

· DOI_CONCEPT_NAME: the string corresponding to the unique concept identifier, from the lookup in the CONCEPT table of the OMOP vocabulary

· HOI_CONCEPT_ID: unique identifier for the outcome, using the OMOP vocabulary

· HOI_CONCEPT_NAME: the string corresponding to the unique HOI concept identifier, from the lookup in the CONCEPT table of the OMOP vocabulary

· GROUND_TRUTH: classification of whether the drug-outcome pair was a positive control (1) or negative control (0).

· ANALYSIS_ID: the unique identifier for the specific parameterization of the method executed

· METHOD_ABBR: abbreviation for the method executed

· RR: effect estimated generated by the method, standardized to the ‘relative risk’ scale

· LB95RR: lower bound of 95% confidence interval around the relative risk estimate

· UB95RR: upper bound of 95% confidence interval around the relative risk estimate

· SIGNIFICANT_RR: binary indicator if effect estimate was statistically significantly positive at p<0.05 (if LB95RR > 1, then SIGNIFICANT_RR = 1, else SIGNIFICANT_RR = 0)

· LB95DIFF: RR – LB95RR

· UB95DIFF: UB95RR - RR

Visualization: Scatterplot

· Data: HOI_score_full

· Appearance: None

· X-axis: RR

· Y-axis: SOURCE_ABBR

· Colors: SIGNIFICANT_RR, GROUND_TRUTH

· Size: None

· Shape: Fixed shape (diamond)

· Labels: None

· Tooltip: SOURCE_ABBR, DOI_CONCEPT_NAME, HOI_CONCEPT_NAME, GROUND_TRUTH, METHOD_ABBR, RR, LB95RR, UB95RR

· Legend: ‘Show legend’ checked. Display ‘Color by’

· Trellis: Rows and columns/ Columns: DOI_CONCEPT_NAME, HOI_CONCEPT_NAME, METHOD_ABBR

· Line Connection: None

· Market By: Row Number

· Lines & Curves: Vertical Line: 1.00; grey, dashed, foreground, 75% transparency

· Error Bars: Horizontal – RR; Upper error: UB95DIFF, Lower error: LB95DIFF; Check ‘Show end caps’ and ‘Include error bars in axis range’