Leveraging healthcare utilization to explore outcomes from musculoskeletal disorders: methodology for defining relevant variables from a health services data repository

The Defense Health Agency (DHA) supports the delivery of healthcare to approximately 10 million beneficiaries of the MHS through the TRICARE Health Plan. The Decision Support Division of the DHA is responsible for managing the specifications of the MHS Data Repository (MDR), which serves as the centralized repository for all DHA corporate healthcare data fed from a worldwide network of more than 260 Department of Defense (DoD) healthcare facilities and a few non-DoD entities. The MDR collects an incredible volume of data from around the world every single day. It captures, archives, validates, and merges data from over three dozen independent healthcare databases, and then quickly integrates and distributes it in a way that can be used for clinical and business decision support. Data capture comes with unique challenges in that it has to account for military healthcare encounters all around the world, to include austere environments during training, combat and humanitarian operations overseas, as well as care that occurs on naval vessels such as aircraft carriers and submarines. This repository includes records of every single person-level interaction for healthcare where the TRICARE Health Plan is the payer, both inpatient and outpatient, and either in a civilian or DoD facility. For all visits within military treatment facilities (MTF), these person-level visits include variables such as vital signs, body mass index, tobacco usage, inpatient and outpatient medications, and chemistry lab results. For records from fiscal year 2000 and forward, the MDR contains a unique person identifier allowing person-level files to be linked across data sources, and is considered the most reliable source for MHS data. It also contains accounting data for each MTF, beneficiary, and staffing. The MDR data dictionary is publicly available and can be accessed at the DHA’s health.​mil website. The structure of the data is Serial-Attached Small Computer System Interface.

Over 38 unique data files from hundreds of data sources feed into the MDR daily from around the world, and files from each source are aggregated by fiscal year. Data across multiple files are connected based on a unique person identifier. The following are the data files from the MDR we found most relevant to the study of musculoskeletal conditions, and stored in the MDR in Statistical Analysis System (SAS) format (except for pharmacy data, which is stored in text format):

Data are pulled across year and across files through linking of unique person identifiers common to all files. This allows for longitudinal assessment of a large variety of healthcare variables at the single person-level. Utilization of data for research purposes require several steps. First a Data Sharing Agreement must be approved by the DHA Privacy Board, and second an analyst with proper training and access to the MDR must be identified.

The DHA utilizes a robust method for addressing errors in data. Data from MTFs are transmitted from the electronic medical records to the MDR daily, and the MDR processes the data weekly. The data at this point are raw (unprocessed). The master data file is updated monthly, and at this point, both new processed data and updates to previous data are added to the master file. Requests for healthcare utilization through a given time period wait a minimum of 90 days after the event of interest in order to ensure the data are captured, processed, and updated in the master data file. Any files containing a blank encounter identification number, International Classification of Diseases (ICD) or Current Procedural Terminology (CPT) code, or with a visit count of zero, are written to an error file. The exception are CAPER files that are not required to have a CPT code for every visit. These encounters continue to go through an extensive validation process across multiple sources to in order to fill in the missing variables. This process is explained in detail via documentation publically available online [17]. Once an encounter is validated (not raw), it will be uploaded into the master file in a subsequent update.

For purchased care occurring outside of a MTF, the claims are initially submitted by the Managed Care Support Contractors for payment by TRICARE. After the payment has been processed, the records are uploaded to TRICARE’s TED system on a monthly basis, and then processed and stored as a final dataset in the MDR. Data from the MDR are added to the TED encounters as they are processed to improve the utility of the records.

Using healthcare utilization to determine study eligibility comes with several challenges. As with similar planning for other study designs, homogeneity of subjects is important in order to derive generalizable conclusions. While there are many factors to consider, three are key when initially creating a cohort from a healthcare database: inclusion criteria, exclusion criteria, and database eligibility. The use of ICD codes are associated with every encounter and enable filtering of care based on diagnosis. Finally, confirmation that patients are eligible beneficiaries during the entire period of surveillance is important. In the MDR, this is done through the Defense Enrollment Eligibility and Reporting System (DEERS) file.

The following is a description of the methodologies utilized in the data extraction, derivation, and definition of variables from the MDR. The variables represent healthcare events captured during a 36-month period of time for patients undergoing hip surgery.

For our analyses, we targeted all patients ages 18 to 50 who had undergone hip surgery for femoroacetabular impingement (FAI). Because a dedicated diagnosis code for FAI does not exist, we identified common surgical procedure codes for this condition as inclusion variables (Table 1). We then excluded any non-FAI conditions that might also receive this same surgical procedure (Table 2). We did not exclude persons with these codes that occurred after the surgery, as they may have had a relationship to the surgical procedure. Finally, we had to exclude everyone that was not continuously eligible as a TRICARE beneficiary for the entire 12 months before and 24 months after the surgical procedure (Fig. 1).

One of the most notable challenges when working with large datasets is the creation of meaningful and appropriately descriptive codes for quantitative use. Codes for timing and types of procedures, provider type, and use of accessory care options (e.g., medications) are also available and essential. These include surgery, radiology, injection, and physical rehabilitation codes. Identifying providers who deliver different types of care can provide information about optimal pathways of care. This was possible because all medical encounters from the MDR data provided the Healthcare Provider Taxonomy Code, as established by the Health Insurance Portability and Accountability Act (HIPAA). This code identifies the provider type, such as “physical therapist,” “nurse practitioner,” “family physician,” etc. These codes can be found on the Centers for Medicare and Medicaid Services government website. The MDR database also provides a product line code, which indicates the service department where care took place (eg. primary care, orthopaedics, physical therapy, etc.) for direct care (CAPERS and SIDR), but is based on the provider specialty for purchased care (TED). All prescriptions provided to patients during the entire period of surveillance were also abstracted, therapeutic class codes for opiate drugs (280808 and 280812) were flagged. The MDR also provided the date and total days’ supply of the prescription.

A key component of health services research is access and timing to care [22]. Because the MHS is a closed single-payer system, and healthcare payments are standardized among recipients, access to care due to personal cost is a lesser issue than in civilian care. Within the MDR, timing of care is available, as every medical visit documented has a date of encounter. Past studies have found that the timing of care can reduce downstream costs and unnecessary procedures [23, 24]. Some providers may also utilize more medical resources than others. For example, utilization of healthcare and associated costs varies widely based on the type of provider seen. Patients with low back pain who seek care from a physical therapist often have fewer radiographs and surgeries [23, 24]. In contrast, those who are seen in physiatry are at an increased risk for radiographs, advanced imaging, invasive injections, surgery and other healthcare costs [25]. Chiropractic care is associated with decreased advanced imaging or surgeon visits, similar to physical therapy, but with an increased length of time in intervention [26]. Some evidence also suggests that seeing multiple providers leads to higher costs of care [23]. Lastly, there are often regional differences in the type of services individuals receive. In the United States for example, the Northeast region has the fewest surgeries and injections, and the Midwest and South have the highest [18, 24].

An essential element of data for use in health services research is the ability to accurately reflect the health condition, at a person-level, for each individual. Historically, researchers have used comorbidities to describe person-level conditions that could mediate outcomes, and control for the influence of these variables on the targeted outcomes during modeling. These are important to consider, as many analyses related to outcomes should adjust for these conditions and their influence on prognosis. Depending on the analysis, conditions such as the presence of malignant neoplasms can be identified; however, in most cases these individuals are excluded from the cohort. We identified a list of medical comorbidities within the MDR that we found to have a significant association with orthopaedic injury and surgical outcomes (Table 3). The MDR validation system also includes identification and flagging of industry-related codes related to patient disease profile. These include the Center for Medicare and Medicaid Services (CMS) Hierarchical Coexisting Conditions (HCC) risk score which calculates a risk score for adjusted comparisons of patients with various disease profiles and the Agency for Healthcare Research and Quality (AHRQ) Chronic Condition Indicator (CCI), which categorizes conditions into chronic and not chronic.

HCCs are proprietary formulas, typically used for Medicare populations, and use data to prospectively estimate predicted costs for enrolled individuals during a follow-up year of coverage. These predictions are based on demographic information and major medical conditions documented from patient encounters in a previous 12-month period. HCCs provide a cost-related risk to the overall outcome since these define which codes are related to the highest costs for care. We used a regression based risk adjustment mechanism to control risks of higher costs based on selected health conditions that is similar to HCC. Our weighted risk adjustment variables were built on the presence of comorbidities and the weighted influence of those comorbidities, in single or in combination, on the outcomes of costs and visits.