Clinical Trial Harmonization

Harmonize your entire trial portfolio into proprietary, analysis-ready data assets

Tremendous value lies dormant in completed clinical trial datasets, but accessing cross-trial insights depends on the ability to accurately and efficiently harmonize trial results to unified SDTM, ADaM, or other formats. Cornerstone’s AI-powered clinical trial harmonization tool accelerates  research by maximizing the use of team's existing internal and external data assets. Leverage pooled datasets to inform future trial design, create benchmarks for monitoring live trials, and more with Cornerstone.

1

Use pooled trial data from live and/or completed studies to train internal AI models

2

Improve new protocol designs using what has worked - and what hasn’t.

3

Turn heterogenous data sources into large-scale, analysis-ready resources

The Challenge

SDTM format is focused on submission to FDA, and while strict in some areas, it allows for flexibility across trials. For example, laboratory results, tumor response, and other key domains are not required to be standardized by the FDA. These domains may not be well standardized within any one trial, and are rarely well-standardized across trials.
‍
As a result, traditional trial harmonization approaches are resource- and time-intensive.

One top-10 sponsor estimated that historically it took 1 - 3 months of technical employee time to manually standardize a single trial. In addition to this direct economic cost, the opportunity cost associated with the time it takes to complete vs. work on other higher-value projects is substantial and often underlooked.

‘Event’ Tables Dictionaries
‘Test’ Tables Dictionaries

‘Event’ Tables Dictionaries

Domain
Ontology
Challenges
Adverse Events
MedDRA (Preferred Term)
Lacking consistency across trials
Concomitant Medications
WHODrug + CDISC Frequency/Route
Lacking consistency across trials
Medical History
MedDRA
Poorly Standardized
Procedures
CPT
Poorly Standardized

‘Event’ Tables Dictionaries

Domain
Ontology
Challenges
Labs
CDISC Lab Test/Unit
CDISC is limited, noncommon labs may not be standardized and require other dictionaries (eg LOINC)
Disposition, Response
Not Standardized to Consistent Ontology
Custom dictionaries are needed for test name, test category, result
Our Product

With Cornerstone, teams can access fully and consistently standardized SDTM, ADaM, or EHR data across multiple study datasets.

Unlock the potential of aggregated data assets

AI-powered platform that ingests, maps, and processes any dataset

Cornerstone can ingest any native SDTM, ADaM, or other data schema and produce consistent outputs across tables and fields.

Add standardizations where none exist across clinical domains

Cornerstone’s NLP allows for automated standardization of critical endpoints that may not be standardized within or across clinical trials, such as clinical lab names, units and values. In addition to supporting industry-standard terminologies, Cornerstone supports custom dictionaries to meet the needs of every dataset.

Validate the consistency of existing standardizations from verbatims

Where fields such as adverse events and concomitant medications, have been previously standardized by each trial team, Cornerstone produces an independent prediction and can validate the original standardization or flag terms which have been inaccurately or inconsistently standardized.

Reduce time to harmonized data by weeks with full human traceability

Human-in-the-loop processing with a dedicated user interface for technical teams to review AI predictions and make additional adjustments if desired. All transformations tracked in automated, exportable audit trail.

Clinical Trial Harmonization In Action

SDTM does not require standardization of many tables and fields, leading to trial-specific naming conventions. Cornerstone adds lab , adverse event, and medication data that is consistent and comprehensive, while maintaining the original data for traceability and auditability.

Incompatible Measurements
Uncoded Terms
Inconsistent Coding
No Standardization
Mislocated Data
The Problem
Test names, units and results vary significantly across clinical trials, making it impossible to confidently group the same tests and measurements across trials.
The Reason
The FDA only requires AE and CM to be standardized. As a result lab tests are standardized to suit a particular study’s objective, but rarely standardized in a consistent way across trials.
The Solution
Cornerstone re-evaluates the original test name, specimen [not shown], unit and result, and assigns standard test names as well as harmonized units.
Before
Study
LBTEST
LBORRES
LBORRESU
1
D-Dimer
3310
ng/ml FEU
2
DDIMER FEU
3.51
ng/ml FEU
3
Fibrin D-Dimer
1280
ng/ml
4
D Dimer
1640
ug/L
5
DDIMER
1.66
ug/mL
6
fibrin ddimer
1.8
ug/L
The original test names, results, and units vary substantially across trials.
With Cornerstone
Study
LBTEST
LBORRES
LBORRESU
1
D Dimer
3310
ng/ml FEU
2
D Dimer
3510
ng/ml FEU
3
D Dimer
2560
ng/ml FEU
4
D Dimer
3280
ng/ml FEU
5
D Dimer
3320
ng/ml FEU
6
D Dimer
3600
ng/ml FEU
Cornerstone harmonizes test results for consistency & comparability.
Formerly
Now
Before
Study
LBTEST
LBORRES
LBORRESU
1
D-Dimer
3310
ng/ml FEU
2
DDIMER FEU
3.51
ng/ml FEU
3
Fibrin D-Dimer
1280
ng/ml
4
D Dimer
1640
ug/L
5
DDIMER
1.66
ug/mL
6
fibrin ddimer
1.8
ug/L
The original test names, results, and units vary substantially across trials.
With Cornerstone
Study
LBTEST
LBORRES
LBORRES
1
D Dimer
3310
ng/ml FEU
2
D Dimer
3510
ng/ml FEU
3
D Dimer
2560
ng/ml FEU
4
D Dimer
3280
ng/ml FEU
5
D Dimer
3320
ng/ml FEU
6
D Dimer
3600
ng/ml FEU
Cornerstone harmonizes test results for consistency & comparability.
The Problem
Adverse events and concomitant medications may not have standard terms or codes present in the study data. In the example below records containing oncology regimen information have not been coded.
The Reason
Coding may be absent due to data format (EDC vs. SDTM), trial status (discontinued or live), and/or because the study team included concepts which require additional processing before being mapped to WHODrug, like in the example below.
The Solution
Cornerstone supports mapping to WHODrug and MedDRA via NLP. In the oncology example below the regimen must first be standardized to the NCI Cancer regimen (see After-CMDETAIL), expanded to a single record per drug, and then standardized to WHODrug.
Before
USUBJID
CMTRT
CMDECOD
1
ACETAMINOPHEN
3310
2
ENOXAPARIN
3.51
3
RCHOP 8 Cycles
1.8
Cancer regimen not in WHODrug and has not been standardized in CMDECOD.
With Cornerstone
USUBJID
CMTRT
CMDETAIL
CMDECOD
1
ACETAMINOPHEN
 
Paracetamol
2
ENOXAPARIN
 
Enoxaparin
3
RCHOP 8 CYCLES
R-CHOP
Rituximab
3
R-CHOP
R-CHOP
Paracetamol
3
R-CHOP
R-CHOP
Doxorubicin
3
R-CHOP
R-CHOP
Vincristine
3
R-CHOP
R-CHOP
Predinisone
Cornerstone standardizes ‘RCHOP 8 CYLES’ to the appropriate term, creates a record per drug in the regimen, and maps the individual drugs to WHO Drug.
Formerly
Now
Before
USUBJID
CMTRT
CMDECOD
1
ACETAMINOPHEN
3310
2
ENOXAPARIN
3.51
3
RCHOP 8 Cycles
1.8
Cancer regimen not in WHODrug and has not been standardized in CMDECOD.
With Cornerstone
USUBJID
CMTRT
CMDETAIL
CMDECOD
1
ACETAMINOPHEN
 
Paracetamol
2
ENOXAPARIN
 
Enoxaparin
3
RCHOP 8 CYCLES
R-CHOP
Rituximab
3
R-CHOP
R-CHOP
Paracetamol
3
R-CHOP
R-CHOP
Doxorubicin
3
R-CHOP
R-CHOP
Vincristine
3
R-CHOP
R-CHOP
Predinisone
Cornerstone standardizes ‘RCHOP 8 CYLES’ to the appropriate term, creates a record per drug in the regimen, and maps the individual drugs to WHO Drug.
The Problem
It is common for the same or clinically close terms to be coded in different ways in different studies, introducing variability in the pooled data despite consistency within each trial.
The Reason
This variation is due to a combination of the complexity of the terminologies and the fact that trials are managed and coded by separate teams with distinct study aims.
The Solution
Cornerstone supports coding to MedDRA and WHO Drug via NLP to ensure that the source values are systematically and consistently coded across trials.
Before
Study
AETERM
AEDECOD
1
TRANSAMENITIS
hypertransaminasaemia
2
TRANSAMENITIS
transaminases increased
3
TRANSAMENITIS
transaminitis
The term ‘transaminitis’ is a lower-level term, and not a preferred term, and transaminases increases is related but not as precise as hypertransaminasemia.
With Cornerstone
Study
AETERM
AEDECOD
1
TRANSAMENITIS
hypertransaminasaemia
2
TRANSAMENITIS
hypertransaminasaemia
3
TRANSAMENITIS
hypertransaminasaemia
Cornerstone produces consistently and accurate standardized terms.
Formerly
Now
Before
Study
AETERM
AEDECOD
1
TRANSAMENITIS
hypertransaminasaemia
2
TRANSAMENITIS
transaminases increased
3
TRANSAMENITIS
transaminitis
The term ‘transaminitis’ is a lower-level term, and not a preferred term, and transaminases increases is related but not as precise as hypertransaminasemia.
With Cornerstone
Study
AETERM
AEDECOD
1
TRANSAMENITIS
hypertransaminasaemia
2
TRANSAMENITIS
hypertransaminasaemia
3
TRANSAMENITIS
hypertransaminasaemia
Cornerstone produces consistently and accurate standardized terms.
The Problem
Clinical trial data contains data with spelling errors, abbreviations, and other sources of variation introduced by clinical sites.
The Reason
It is common for electronic data capture of clinical trial data to allow for free-text entry. While fields supporting safety analysis are likely to be standardized, many other free text field are likely to remain unstandardized by the trial team.
The Solution
Cornerstone supports standardization to CDISC terminology as well as the creation of custom terminologies where standards may not be available.
Before
HOTERM
HODECOD
Study drug therapy
Study drug administration
Study therapy
Study cycle 1 administration
progression
progressive disease
disease progression
The original, verbatim terms contain significant variation but the standard SDTM column has not been populated in the study.
With Cornerstone
HOTERM
HODECOD
Study drug therapy
Study treatment administration
Study drug administration
Study treatment administration
Study therapy
Study treatment administration
Study cycle 1 administration
Study treatment administration
progression
Disease progression
progressive disease
Disease progression
disease progression
Disease progression
Cornerstone produces consistently and accurate standardized terms.
Formerly
Now
Before
HOTERM
HODECOD
Study drug therapy
Study drug administration
Study therapy
Study cycle 1 administration
progression
progressive disease
disease progression
The original, verbatim terms contain significant variation but the standard SDTM column has not been populated in the study.
With Cornerstone
HOTERM
HODECOD
Study drug therapy
Study treatment administration
Study drug administration
Study treatment administration
Study therapy
Study treatment administration
Study cycle 1 administration
Study treatment administration
progression
Disease progression
progressive disease
Disease progression
disease progression
Disease progression
Cornerstone produces consistently and accurate standardized terms.
The Problem
Records can be mislocated in the incorrect domain. In the example below cytokine test results have been included in the LB domain instead of IS.
The Reason
The SDTM criteria change over time and are open to interpretation by different trial teams to suit the needs of the study.
The Solution
Cornerstone systematically scans domains likely to contain mislocated data, such as LB and CM, to identify which records are mislocated and then moves them to the appropriate domain.
Before
LB Domain
LBTEST
LBCAT
LBORRES
LBORRESU
Platelet Count
HEMATOLOGY
410
10^9/L
Interleukin 10
CYTOKINE
5.4
pg/mL
Glucose
CHEMISTRY
102
mg/dL
Interleukin 2
CYTOKINE
15.2
pg/mL
White Blood Cell Count
HEMATOLOGY
3.2
10^9/L
Interleukin 1 Beta
CYTOKINE
22.1
pg/mL
Hemoglobin
HEMATOLOGY
13.8
g/dL
Interleukin 6
CYTOKINE
48.7
pg/mL
Cytokine results have been mistakenly placed within the LB domain.
With Cornerstone
LB Domain
LBTEST
LBCAT
LBORRES
LBORRES
Glucose
CHEMISTRY
102
mg/dL
Hemoglobin
HEMATOLOGY
13.8
g/dL
White Blood Cell Count
HEMATOLOGY
3.2
10^9/L
Platelet Count
HEMATOLOGY
410
10^9/L
IS Domain
LBTEST
LBCAT
LBORRES
LBORRES
Interleukin 2
CYTOKINE
15.2
pg/mL
Interleukin 6
CYTOKINE
48.7
pg/mL
Interleukin 10
CYTOKINE
5.4
pg/mL
Interleukin 1 Beta
CYTOKINE
22.1
pg/mL
Cornerstone identifies mislocated data, creates the IS domain, and relocates the records to the correct domain.
Formerly
Now
Before
LB Domain
LBTEST
LBCAT
LBORRES
LBORRESU
Platelet Count
HEMATOLOGY
410
10^9/L
Interleukin 10
CYTOKINE
5.4
pg/mL
Glucose
CHEMISTRY
102
mg/dL
Interleukin 2
CYTOKINE
15.2
pg/mL
White Blood Cell Count
HEMATOLOGY
3.2
10^9/L
Interleukin 1 Beta
CYTOKINE
22.1
pg/mL
Hemoglobin
HEMATOLOGY
13.8
g/dL
Interleukin 6
CYTOKINE
48.7
pg/mL
Cytokine results have been mistakenly placed within the LB domain.
With Cornerstone
LB Domain
LBTEST
LBCAT
LBORRES
LBORRES
Glucose
CHEMISTRY
102
mg/dL
Hemoglobin
HEMATOLOGY
13.8
g/dL
White Blood Cell Count
HEMATOLOGY
3.2
10^9/L
Platelet Count
HEMATOLOGY
410
10^9/L
IS Domain
LBTEST
LBCAT
LBORRES
LBORRES
Interleukin 2
CYTOKINE
15.2
pg/mL
Interleukin 6
CYTOKINE
48.7
pg/mL
Interleukin 10
CYTOKINE
5.4
pg/mL
Interleukin 1 Beta
CYTOKINE
22.1
pg/mL
Cornerstone identifies mislocated data, creates the IS domain, and relocates the records to the correct domain.

Case Study

Customer

Top 10 Pharma company

The Need

The company indicated that an optimal solution would help them save time by reaching high thresholds of standardization and accuracy for automatically assigned mappings across SDTM tables. In addition, they wanted an intuitive user interface (UI) to allow for straightforward expert review of the output.

Impact

Cornerstone AI capabilities evaluated on three blood cancer clinical trials, completing the pilot in a matter of days.

‘Event’ Tables Dictionaries
Domain
Sponsors' Success Criteria
Results Across 3 Trials
Standardize Fields Using Raw Terms
  • >75% auto-standardization rate
  • >90% accuracy rate for auto-standardizations (assessed via comparison to original study standardized terms + manual adjudication)
  • Intuitive and efficient UI for manual standardization
  • 93-99% auto-standardization rate
  • 98-99% accuracy rate
  • Custom dictionaries created
  • Intuitive and efficient UI for manual standardization
Harmonize Numeric Results
  • Consistent naming, units, values
  • 99% test name consistency across studies
  • 100% accuracy in unit harmonization
Reduce Overall Time to Harmonize Clinical Trials
  • Complete within ~6 weeks (the company’s internal harmonization benchmark)
  • Completed in 3 days

Turn disparate trial datasets into aggregated, AI-ready assets today

Contact Us