Outcomes Research Programming vs Traditional Clinical Trial Programming

Outcomes research aka health outcomes research, is the study of the end results of particular health care practices and interventions, in other words it is the study of what happens in the real world to patients when they are given a certain treatment or a certain method of care. Outcomes research studies are used to improve the quality and value of healthcare for patients.

Longitudinal Databases

To assess what is happening in the real world, rather than using clinical trials to collect data, outcomes research often uses retrospective, non-interventional studies performed on huge databases which contain de-identified medical records. These databases are often called longitudinal databases as they are able to track patients over multiple years. Each database has its own structure, advantages and limitations. There are a few possible data sources:

• Administrative medical and pharmacy claims, sourced from public and private insurance companies:
  • Truven Health MarketScan® Research Database the largest claims database available for licensing, with more than 170 million unique patients since 1995. Contains fully integrated patient-level data (inpatient, outpatient, drug, laboratory, health and productivity management, health risk assessment, dental, and benefit design) from commercial insurance, Medicare supplemental claims, and Medicaid claims to reflect real-world treatment patterns and costs
  • Clinformatics Data Mart (An Optum company), which includes enrolment information and administrative health claims from 1993 to present day on more than 114 million lives
• Clinical databases - systematized collection of patients’ electronically-stored health information, sourced from GPs, hospitals:
  • The Health Improvement Network (THIN) Database is a large UK primary care database. The THIN Database Research Team uses this database, containing data collected from over 550 general practitioners spread over the UK, for research into cardiovascular disease, mental health, pharmacoepidemiology and other fields of primary care research
  • Premier healthcare Database – which provides broad and detailed insights from over 700 U.S. hospitals with comprehensive billing, cost, device, medication and procedure data
• Survey data
  • The National Health and Wellness Survey
  • The National Health Interview Survey
  • The National Health and Nutrition Examination Survey

All analysed databases are updated on a regular basis (varying from 4 times per year to annual basis), allowing for investigation of trends in treatment pattern using current data.

Working with Big Data & Teradata SQL

Due to their huge size and differing structures, working with these databases presents many challenges.  In particular, data extraction times and the resulting extracted datasets can become unmanageably large with traditional SAS programming techniques.

Therefore it requires an appropriate environment which would make programming  on these large datasets more efficient.  Raw data obtained from vendors are pre-processed and then loaded into local systems (data warehouse) designed to manage big data.

At Quanticate we have been using the Teradata database and find it to be more efficient than the standard SAS datasets as it is designed in a way to handle the large volumes of big data compared to typical SAS data structure.

A programmer needs to demonstrate a technical proficiency of handling large datasets and be familiar with Teradata SQL syntax in order to process the data directly into the database. This way, handling huge datasets is less time consuming. Only final extracts are downloaded into SAS datasets for further analysis.  Other common companies providing the required environment for programming big data are Oracle, IBM and Microsoft. 

Outcomes Research Study Design

When programming an outcomes research project, a sponsor may be looking for experienced programming resource with an understanding of statistics and database design. Data needs to be handled carefully, keeping in mind all limitations. Outcomes Research programmers provide all required information and advise on best approaches to reach the realisation of the client request.

Data Usage

Instead of the typical clinical trial structure where you are testing against a predicted hypothesis to validate the efficacy of a drug, Outcomes Research study design requires less specific terms and a more general approach that looks to monitor the real world data to find patterns and health outcomes. You can perform this type of research even before a clinical trial has started to assess if the potential study population is suitable for the desired trial.

An Outcomes Research study can consist of any number of the following parts:      

• Measuring presence of a disease in a population,
• Tracking usage of given drug among selected population across years,
• Comparing selected drugs,
• Analysing Health Care Resource Utilizations and costs,
• Analysing potential clinical trial population,
• Comparing case and control cohorts,
• Analysing treatment patterns (persistence on given drug, switches to competitors, uses of generics),
• Conducting sponsor’s specific analysis.

Requested analysis depends on the study design. In addition to descriptive statistics, more sophisticated analysis can be performed (i.e. logistic regressions and survival analysis).

Three Phases of Outcomes Research Studies

Typical Outcomes Research studies consist of 3 phases:

1. Defining/extracting cohort (e.g. list of patients meeting inclusion/exclusion criteria).
2. Performing requested analyses for selected patients. Most common are: demographic characteristics, comorbidities analysis, drug persistence, cost and HCRU.
3. Extracting required data for the patients in cohort at the request of the sponsor’s hypothesis: inpatient/outpatient claims, drug claims etc.

Usually there is one main programmer and one QCer which means that a single programmer is independently taking care of all the phases. For more complicated studies, input from the main programmer has an impact on the final shape of the study. It requires close cooperation with the sponsor and statistician.