Databases for Gastrointestinal Clinical and Public Health Research: Have Database, Will Research

Graphical Abstract

So, you want to do GI clinical research but aren’t sure how to get started? Although randomized clinical trials provide the highest level of evidence to inform clinical practice, guidelines, and policy, they may not be possible for a fellow or junior investigator to initiate, and even in the best-case scenario, may not be published for many years. Database-oriented research can overcome many of the daunting impediments that many junior researchers face. Furthermore it can jump-start a clinical research career as a vehicle to develop knowledge and experience in epidemiologic analyses and publication. It offers a way to test clinically meaningful hypotheses and most importantly, provide salient data that could ultimately impact clinical decision making.

What are the next steps to engaging in database research? Finding a mentor with experience in clinical and database research is critical. If necessary, mentorship can be split between a gastrointestinal (GI) clinical researcher who can help identify clinical questions and another mentor with experience in epidemiologic analyses (eg, health services researcher, epidemiologist, and biostatistician). The next step is to identify the clinical area within GI that is of most interest to you and possibly your future clinical focus (eg, cancer screening and prevention, motility, hepatology, inflammatory bowel disease [IBD]).

Although ideally you would have a hypothesis to be tested, in database-oriented research, it can be advantageous to explore the specific database to see what hypotheses can be investigated, using an iterative approach to home in on a meaningful analysis. Because of this low barrier to entry however, a thorough literature review is important because many of the analyses that you may consider may have already been done. Thus, a willingness to be flexible is critical to success.

In this piece, we provide synoptic review of numerous databases suited to GI-focused analyses (Table 1). The goal is not necessarily to be comprehensive but focused and provide illustrative examples with the ultimate aim to point an eager GI fellow or junior investigator in the right direction.

Table 1.

Review of Databases

Database name	Access	Cost	Type of database	Data included	Why you would consider using this
NHANES https://www.cdc.gov/nchs/nhanes/index.htm	Publicly available	Free	National Survey	Interview obtained medical information, physical exam and laboratory data	Investigate diseases and associations with medications, physical exam, and laboratory data
National Health Interview Survey https://www.cdc.gov/nchs/nhis/index.htm	Publicly available	Free	National Survey	Interview obtained medical information	Investigate association between diseases and health care access and barriers
SEER/ SEER Research Plus https://seer.cancer.gov/	Publicly available For certain restricted data an application is required.	Free	Registry	Patient demographics and cancer specific data (such as primary tumor site, morphology, and stage)	Asses cancer incidence and mortality, survival, and limited duration prevalence
GI Quality Improvement Consortium (GIQUIC) https://giquic.gi.org/	Application process	Fee	Registry	Endoscopy center and provider characteristics, patient demographics, endoscopy reports and quality metrics	Describe endoscopy and colonoscopy measures
Veterans Affairs https://www.va.gov/vetdata/	Affiliation required	Free	Electronic health records	Entire medical charts	Evaluate longitudinal care of large cohort of patients
Marketscan https://www.ibm.com/products/marketscan-research-databases	Publicly available	Fee	Claims data	Insurance claims, patient demographics. Supplements include laboratory data, disability data, weather data and others	Examine treatment patterns, patient adherence, and natural history of disease across both inpatient and outpatient care
SEER-MEDICARE https://healthcaredelivery.cancer.gov/seermedicare/	Application process	Fee	Combined registry and claims data	Includes SEER registry data along with inpatient, outpatient and medication claims data	Perform epidemiological and health services research in patients with cancer
Nationwide Inpatient Sample https://www.hcup-us.ahrq.gov/nisoverview.jsp	Application process	Fee (discounts available for students)	Combined registry and claims data	Claims data, patient and provider characteristics	Evaluate inpatient care
All of Us https://allofus.nih.gov/	Two access types: anonymized aggregate data publicly available and more comprehensive individual data for registered researchers	Free	Prospectively enrolling registry, including survey data, electronic health records, wearable device data and biosamples; data are both prospective and retrospective	Patient health information from survey and electronic health records, biosamples	Evaluate the health of a large, diverse cohort of patients, including access to biosamples

Types of Health Research Databases

Survey Data

There are many survey databases obtained through participant interview that are readily available to researchers. Two examples are the National Health Interview Survey (NHIS) and the National Health and Nutrition Survey (NHANES), which are conducted by the National Center for Health Statistics (NCHS).^1,2 Both are large national surveys, with NHANES including 5000 individual people per year and NHIS including 30,000 households per year. While both surveys collect sociodemographic, health, and disease information, NHIS is solely interview based and is focused on health care utilization and access. A recent study used NHIS to evaluate food insecurity, social support, and financial toxicity in patients with IBD.³ In addition, every 5 years, NHIS includes cancer-related questions,⁴ which can be used to study cancer screening and risk factors, such as a recent study assessing adherence to colorectal cancer screening guidelines in African Americans.⁵

NHANES, on the other hand, also includes data from physical examination and laboratory testing. Kim et al⁶ used NHANES to show that reduced thyroid function predicted mortality in patients with nonalcoholic fatty liver disease. Despite the advantages, these surveys may be difficult to navigate without a knowledge of statistics. To make the sample population’s answers to survey questions representative of a larger population, the samples need to be weighted, or corrected, by demographic characteristics to improve the accuracy of survey estimates.

Registries

There are also databases of national registries. A registry collects detailed information about a set of patients, such as their age, race and ethnicity, sex, diagnosis, and treatments. One example is the Surveillance, Epidemiology, and End Results (SEER) program, which coalesces data from cancer registries and is funded by the National Cancer Institute (NCI).⁷ SEER includes patient demographics as well as cancer incidence, mortality, tumor stage, and morphology. For example, Hur et al⁸ found an increase in esophageal adenocarcinoma incidence and mortality rates from 1975 to 2009.

To perform analyses, researchers must use the free software, SEER*STAT. While this may be a barrier, there are free tutorials available on the SEER website and a robust and responsive helpline. There are also software applications that use SEER*STAT output and expand the type of statistical tests that can be done, such as JoinPoint for trend analysis. While SEER is publicly available, certain demographic and cancer data is restricted and is only available via SEER Research Plus, which requires an application.⁷

Another registry is the GI Quality Improvement Consortium (GIQuIC), which is intended as a repository for quality improvement measures.⁹ This registry includes patient demographics, provider characteristics, American Society of Anesthesiologists status, anticoagulation use, and endoscopic procedure measures beginning in 2010 from provider practices, ambulatory surgical centers, endoscopy suites, and hospitals. A recent study used this database to describe polyps and neoplasia in patients aged 45 to 49, which in light of recently changing colon cancer screening guidelines is important to informing adenoma detection rates for this age group.^10,11

Electronic Health Records

Electronic health records can serve as a research tool that has the benefit of providing a more complete representation of patient care. Records can be obtained from local hospital systems or from large health systems such as Kaiser or Gei-singer. The benefit of local data is it may be accessible from your own institution, and the patient population will be familiar to you. The downsides of electronic health record data, however, are that the results may not be generalizable outside of a particular geographic area, and the data may be hard to access and may not be immediately user-friendly. Additionally, health records are by definition retrospective and only allow for observational data collection.

One notable example is the Veteran Health Affairs (VA) records, which is a large database of electronic health records that can be accessed with collaboration with VA staff. The VA record has been in place for several decades, providing a unique longitudinal data set. The records include diagnostic and procedural codes, laboratory data, vital signs, imaging, and pathology. One example is a study published in Hepatology in 2021 that described clinical characteristics and outcomes of veterans with and without cirrhosis who tested positive for severe acute respiratory syndrome coronavirus 2.¹² Because of the large population, the authors were able to quickly publish a study at the start of the pandemic including 3306 patients with cirrhosis of 88,747 tested for coronavirus disease. However, because the veterans are often English-speaking and male, findings from these studies may not be generalizable,¹³ and information may be missing if care was provided outside of the VA system.

Claims Data

Data from insurance claims are a robust means of conducting research and can provide information about how care is provided “in the real world.” Data often include patient and provider information, diagnostic and procedural codes, and costs of care. These data sets often lack clinically important information such as laboratory data or social and family history. Another limitation is that it is beholden to provider reporting, which may not be entirely accurate, although it tends to be accurate for procedures (eg, endoscopy and colonoscopy).^14,15

One example is MarketScan, which is a family of databases that include insurance claims from participating providers for their employees and employee dependents. Specifically, the claims data comes from employer-sponsored insurance, employer-sponsored Medicare supplement, and Medicaid in 11 states for inpatient, outpatient, and prescription drug claims, as well as expenditure data. Supplemental data include workplace and disability measures, weather pattern, benefit plan design, and inpatient drug use. Kulaylat et al¹⁶ used the MarketScan database to evaluate postoperative complications of patients with ulcerative colitis who received preoperative anti-tumor necrosis factor therapy.¹⁶ Some limitations of this particular data set are incomplete follow up if patients change employers and the inclusion of only employed patients who are working age.

Another notable database is the SEER-Medicare linkage, which combines cancer data from SEER with Medicare claims.¹⁷ This linkage allows for cancer research that includes measures of comorbidities, receipt of screening and evaluation tests, and detailed treatment data. For example, Rustgi et al¹⁸ found a rise in use of endoscopic ultrasound to diagnose patients with pancreatic cancer during 2000 through 2015 as well as a survival benefit. SEER-Medicare also provides a random 5% sample of patients without cancer to serve as controls so that researchers can conduct population-based analyses within the SEER registry areas.¹⁷ Specific limitations include inclusion of only older patients, cost of data purchase, and an extensive application and data use agreement that needs to be approved by NCI. Unless there is a researcher who already has the database, the time from application to data in hand can be many months.

Finally the National Inpatient Sample is a comprehensive database consisting of national and state-specific data on inpatient stays, ambulatory surgery, and readmissions.¹⁹ It is the largest publicly available all-payer inpatient database designed to produce estimates of inpatient use, access, cost, quality, and outcomes and represents 7 million inpatient stays per year or 35 million per year with weighting. It includes patient demographic and hospital characteristic, International Classification of Diseases codes, charges, discharge status, length of stay, and severity and comorbidity measures. This database was used by Joo et al²⁰ to describe hospital costs and use of palliative care consults and procedures for patients with gastric cancer.

Comprehensive Data Sets

The All of Us Registry is a unique database that is a rich source of data for researchers.²¹ This prospective registry includes survey data at the time of entry and biosamples for future genomic studies and is also linked to health records for a subset of patients. The goal of this registry is to promote health equity, improve wellness and health outcomes, and inform earlier diagnosis of disease in a diverse population of patients. A published study by Renedo et al²² used novel definitions of underrepresented groups, besides race and ethnicity, to demonstrate disparities in revascularization after stroke. This registry may be subject to bias as all participants volunteer to participate.

Conclusion

Our review of the many types of databases that can be used for GI-focused research will hopefully provide the reader with some ideas and help “jump-start” a clinical research career. Although some of these databases can cost as much as several thousand dollars, others are free. Collaboration with other fellows and junior faculty can offset the cost and lead to more fruitful research endeavors.