Conducting multi-institutional clinical research requires a research platform that allows participating research partners to use the data resources in a predictable and efficient way. Common data models are one approach to achieving this aim. PEDSnet makes data available in two common data models, the PEDSnet CDM and the PCORnet CDM.
PEDSnet Common Data Model
chose from the onset of the network to establish a pediatric-specific
Common Data Model (PEDSnet CDM) for the storage of PEDSnet data. The use
of an internal CDM allows PEDSnet to quickly add
data domains or data elements needed by pediatric investigators. Two
examples are age-normalized anthropomorphic measurements (height, weight, BMI
percentiles based on CDC routines) and census block geocoding for
location-based data queries linked to geocoded environmental data
sets. The PEDSnet CDM is based on the Observational Health Data Sciences and Informatics collaborative's OMOP common data model. This CDM focuses strongly on terminology standardization, resulting in
use of common standard terminologies such as SNOMED-CT,
RxNorm, CPT, and LOINC for both clinical and demographic facts. The OMOP model was expanded to include
the PCORnet and pediatric specific data standards, as developed by
PEDSnet. The PEDSnet CDM subsumes all PCORnet elements, but addresses important data elements not yet addressed in the PCORnet-wide CDM.
Because PEDSnet is a Clinical Data Research Network, meaning that its data members are health systems, the majority of the information in the PEDSnet core data resource is related to the delivery of clinical care:
As this spider-web shows, the PEDSnet CDM is anchored by persons and their encounters with either the health system or research studies. Protocol-specific data, such as survey responses or biospecimen status, is also incorporated into the CDM and transmitted to the DCC. This also influences the way each type of information is collected, as this slightly more detailed description of the core domains above and additional components of the CDM highlights:
If you'd like a really detailed view of the CDM, such as you'd need to design a study, please take a look at the CDM ETL Specifications. You're also welcome to contact the Data Coordinating Center with any questions, or to obtain machine-friendly versions of the data model specification. The DCC conducts extensive data quality and characterization analyses, to assess the behavior of data collected in the CDM.
PEDSnet Data Terminology
PEDSnet uses SNOMED-CT as the data terminology for our CDM. Investigators who have ICD-9 codes can visit the following website to explore SNOMED Concepts. Any questions can also be directed to the Data Coordinating Center, where staff will be happy to assist you in finding the right diagnostic codes for your project.
Below are the list and descriptions of current domains in the PEDSnet Database; you may also download the PDF.
|Domain||Examples||Utility to Research|
|Demographics||Age, date of birth, gestational age, sex, ethnicity, race, zip code, PEDSnet health system site, primary care provider (PCP), death and cause of death (if available), tobacco use||Exposures, health determinants, confounders, mediators, effect modifiers, outcomes (death); Zip codes and health system site can be geocoded and linked to area-level environmental factors influencing health and health outcomes;|
|Outpatient encounters (~75 visit specialty types available)||Primary care visits, specialty care clinics (e.g., cardiology, endocrinology, nephrology, oncology, etc.), physical therapy, occupational therapy, speech language pathology, medical genetics/genomics, etc.||Provides longitudinal follow-up and duration of clinical course; health care utilization measures; Distinguishes between specialty care (and type of specialty care) and primary care visits;|
|Inpatient admissions||Length of stay, discharge status; diagnoses, procedures, medications, and lab results associated with inpatient stay||Measures health care utilization for patients and conditions; Provides longitudinal data to evaluate disease course and severity of illness;|
|Emergency department encounters||Diagnoses, procedures, medications, and lab results associated with ED visit; ED visits resulting in inpatient admission||Provides markers for disease progression and clinical outcomes; Health care utilization measures; Markers for chronic disease control and severity;|
|Anthropometrics||Height (cm) and weight (kg), BMI, head circumference||Outcomes, health determinants, confounders, mediators, effect modifiers; Markers for health status;|
|Vital signs||Temperature, blood pressure||Outcomes, health determinants, confounders, mediators, effect modifiers; Markers for health status|
|Providers||Specialty, health care facility||Differences and similarities in clinical course of care between patients seen by primary care and specialty care physicians;|
|Diagnoses||Final diagnoses are recorded at each encounter and are mapped to a standardized codes (typically SNOMED-CT)||Provides ability to define and construct cohorts and outcomes of interest; Sufficient power to study rare diseases of interest; Provides comorbidity information;|
|Procedures||Procedures are recorded at each encounter and are mapped to standardized SNOMED-CT, ICD-9 Procedure, ICD-10 Procedure, CPT-4, and HCPCS codes||Measures disease severity; Improves case-finding accuracy and provides more detailed clinical data;|
|Prescribed medications||Medications ordered are recorded at each encounter and mapped to a standardized RxNorm code||Provides information about disease control and severity; Markers of clinical course and health behaviors;|
|Dispensed medications||Medications dispensed, when available, are recorded and mapped to a standardized RxNorm code||Medications that patients likely received; More accurate measure of medication record and disease control|
|Laboratory test results||Key components of lipid panel, complete metabolic panel, complete blood counts, microbiologic cultures, liver function tests, urinalysis, viral panels, etc.||Identifies more precise clinical outcomes and health determinants; Evaluate and define disease progression; Provides more accurate measure and definition of disease as well as disease severity and control;|
|Visit Payer||Plan class (private/commercial, medicaid/sCHIP, Medicare, other public, self-pay, other/unknown), plan type (HMO, PPO, POS, fee for service, other/unknown) for every unique visit. Enrollment information unavailable.||Confounders, mediators, effect modifiers; Ability to study effect of payer visit with health outcomes and other outcomes of interest|
The Demographics Domain contains demographic data at the person-level for all patients in the PEDSnet database. Example variables include age, date of birth, gestational age, sex, ethnicity, race, zip code, PEDSnet health system site, primary care provider (PCP), death and cause of death (if available), and tobacco use. Demographic data is a key element in all clinical health research and used to measure exposures, health determinants, confounders, mediators, effect modifiers, and health outcomes (e.g., death or cause of death). Researchers can also geocode zip codes and health system sites and study the link between environmental factors and health outcomes.
The Outpatient Encounters Domain captures all outpatient encounters for patients in our database who attend a site-affiliated office or institution. The PEDSnet database identifies ~75 different visit specialty types. Examples of specialty visits include primary care offices, specialty clinics (e.g., cardiology, endocrinology, nephrology, oncology, etc), development and behavioral appointments (e.g., physical therapy, occupational therapy, speech language pathology), and other miscellaneous visits (e.g., medical genetics/genomics). The Outpatient Encounters Domain provides researchers with longitudinal follow-up and duration of a patient’s clinical course of care as well as health care utilization measures. Information in this domain includes dates of service, diagnoses, procedures, and information about physician and location specialty.
Similar to the Outpatient Encounters Domain, the Inpatient Encounters Domain encompasses all inpatient encounters for patients in our database who attend a PEDSnet site-affiliated hospital. Examples of data included in this domain are length of stay and discharge status, and all diagnoses, procedures, medications, and lab results linked to a unique inpatient visit. This domain is of high utility to researchers. Inpatient medical encounters can serve as measures for healthcare utilization and studied by disease type (e.g., oncology patients) or patient profile (e.g., patients in geographic locations or age groups). The Inpatient Encounters Domain also provides researchers with longitudinal data to evaluate disease course and severity of illness. The application of this domain can extend to almost any type of epidemiologic or clinical health study.
Analogous to the Outpatient Encounters Domain and the Inpatient Encounters Domain, the Emergency Department Encounters Domain consists of all diagnoses, procedures, medications, and labs associated with an ED visit. Similar measures of health care utilization and clinical course of care can be applied to this domain. Additionally, we can readily identify emergency visit encounters that result in inpatient admissions in the PEDSnet database.
The Anthropometrics Domain contains patient-level information regarding height (cm) and weight (kg) and head circumference. This domain serves a similar function as the Vital Signs Domain, which encompasses all temperature and blood pressure measurements for patients in the PEDSnet database. Together, these domains provide researchers with data for outcomes and health determinants in a study, as well as for confounders, mediators, and effect modifiers in statistical analyses. The variables in these two domains serve as markers of general health status.
The Providers Domains encompasses person-level information for any health care professional who provides care for patients enrolled in the PEDSnet database. This domain contains information regarding provider specialty as well as associated health care facility. The data can serve a variety of functions in health care research, such as examining the differences and similarities in the clinical course of care between patients seen primarily by their primary care doctors versus those seen by specialty care physicians. It also provides the ability to flag visits and profile patient characteristics based on specialty care. There are currently > 120 unique physician specialty types encompassed in the database.
The Diagnoses Domain encompasses all diagnostic information for patients in the database (>32,000 unique conditions captured). Final diagnoses are recorded at each encounter and mapped to a standardized code, typically SNOMED-CT. Because diagnostic codes are captured at every encounter, this domain provides researchers with the unique ability to longitudinally assess the duration of a patient’s illness. Almost all health care research hinges on the ability to accurately assess patient diagnoses and duration of illness. The diagnoses domain provides researchers with the ability to define and construct cohorts and outcomes of interest as well as accurately assess comorbidities. Because of its size and the high number of patients enrolled in our database, studies will be sufficiently powered to study rare diseases (e.g., oncology, sickle cell disease, etc) where other databases are too small to provide meaningful data.
The Procedures Domain contains data for all procedures performed on a patient enrolled in the PEDSnet database (>12,500 procedures captured). Similar to the diagnoses, procedures are recorded at each encounter and are mapped to standardized codes, which include SNOMED-CT, ICD-9 procedures, ICD-10 procedures, CPT-4, and HCPCS codes. Procedures are a useful measure for researchers and are often used in conjunction with diagnostic codes to improve case-finding accuracy and provide more detailed clinical data. It can also be used as a proxy for disease severity or clinical indications.
The PEDSnet database differentiates between the Prescribed Medications Domain and the Dispensed Medications Domain. Both domains capture recorded medications for a given encounter and map to a standardized RxNorm code. However, the dispensed medications domain provides more accurate information about the drugs that the patient actually received and can therefore more accurately assess a patient’s medication record and disease control status. Both domains are vital in the conduct of epidemiologic research and are often used to construct a cohort of interest, assess disease status, define a patient’s clinical course and health behaviors, and measure outcomes of interest.
The Laboratory Test Results Domain contains the orders and results for more than 400 laboratory tests. Examples include key components of a lipid panel, complete metabolic panel, complete blood counts, microbiologic cultures, liver function tests, urinalyses, viral panels, etc. This domain identifies more precise clinical outcomes and health determinants for patients. Researchers can evaluate and define disease progression more easily and measure and define disease status, severity, and control.
The Visit Payer Domain provides plan class (private/commercial, Medicaid/sCHIP, Medicare, other public, self-pay, other/unknown) and plan type (HMO, PPO, POS, fee-for-service, other/unkown) for every unique visit in the database. Researchers can use this domain to study the effect of payer visit type with health (and other) outcomes of interest. Data in this domain can also be used during statistical analyses as confounders, mediators, or effect modifiers to the association of interest.
PCORnet Common Data Model
PCORnet created the PCORnet Common Data Model, summarized below for networks in the PCORnet initiative, to facilitate the sharing of information across the wider PCORnet network. The PCORnet Common Data Model is based on the Mini-Sentinel Common Data Model, and is used for PCORnet sponsored studies and data queries. As a PCORnet network, PEDSnet participates in these queries when approved by PEDSnet governing bodies.
The PEDSnet DCC has developed automated transformations from the PEDSnet CDM to the PCORnet CDM, demonstrating the feasibility of interconversion between CDM representations of similar content, and supporting participation in many collaborative research communities.