H

Approach to Data Analysis and Code Library

INTRODUCTION

The following describes the approach to the data analysis conducted by the committee for the following studies: Study 1 (opioid initiation among those with and without current benzodiazepine fill), studies 2a and 2b (opioid baseline, dose escalation), and study 3 (benzodiazepine initiation among those on long-term opioid pharmacotherapy). Westat, the subcontractor, had access to the data and analyzed the data available through outlined sources. All analyses were conducted using SAS 9.4 and SAS Enterprise Guide.

The following describes the data sources used for the analyses, including covariates, and overall approach to constructing the analytic data files.

Data Sources

Per data use agreements with Veterans Health Administration (VHA), Westat obtained access to health record data from VHA CDW and Centers for Medicare & Medicaid Services (CMS) Medicare files. VHA CDW provided a snapshot of medical records in the VA Informatics and Computing Infrastructure (VINCI) workspace, and the database tables within VINCI were linked using unique individual identifiers assigned by the VHA (PatientICN). CMS datafiles were provided in the database as SAS datasets and linked to VHA medical records by Social Security number (SSN). Death data for SSNs in the VINCI extract were obtained from National Death Index files per an internal request to the Joint Department of Veterans Affairs (VA) and Department of Defense Mortality Data Repository Data team and added back to the veteran record using the SSN.

Only the subcontractor had access to personally identifiable health information and personally identifiable information; all access to the data and analyses occurred within the VA firewall and data ecosystems. All analyses were also conducted within the VA firewall, and the final analytic aggregated results were removed from VINCI to provide to the committee only after privacy review. All approved data analysts who worked on this study were also subject to a VA background check, maintained up-to-date human subjects training, and worked within Westat’s designated workspace on the server. The National Academies of Sciences, Engineering, and Medicine (National Academies) Institutional Review Board approved the study. Study results do not report data if sample sizes were 10 or fewer individuals. This helps to ensure confidentiality of individuals in the study and is in line with CMS data

policy and the National Center for Health Statistics (NCHS) recommendations.^1,2 Furthermore, cells are marked as unreliable where sample sizes were 20 or fewer.

Covariates

Prescription data were obtained from both VHA outpatient prescriptions and CMS Medicare provided Part D claims. Drug list lookup tables were used to identify valid opioid prescriptions that were filled (picked up) from the VHA and CMS.

All diagnostic specific covariates were derived using a valid ICD-9 or ICD-10 code from either VHA CDW or CMS files. For example, veterans with an end stage renal disease (ESRD) code in VHA files, CMS files, or both would be coded as having ESRD.

Demographic variables were obtained from VHA. For some demographics, such as age and marital status, the record closest to the veteran study start date was used. Other demographics were based on all records such as sex, race, and ethnicity.

Overall Approach to Analytic Files

Westat obtained records for any patients with an encounter in the VHA between 2005 and 2020. Data extraction, preparation, cleaning, and transformation of the billions of VHA records was necessary to eventually create the final study analysis datasets.

The first data prep step was to identify eligible veterans for the study (i.e., all study groups). Westat applied inclusion and exclusion criteria approved by the National Academies committee along with VHA best practice to drop suspect records (e.g., test cases and bad SSNs). The next step involved identifying veterans prescribed the main drugs of interest at any time, which involved a manual step of reviewing all the VHA drug codes to flag the drugs of interest and then linking that reference file back to the VHA prescription data.

After identifying eligible veterans and an extract on their drug use, a sample frame was created for each of the studies based on the committee criteria (e.g., no drug use before first study month). A veteran may appear multiple times in the sample frame with different episodes of drug use. For study 1, a random episode is retained for each veteran. For study 2, the first episode is retained for each veteran. For study 3, all veteran episodes are retained, given the much smaller sample size for analysis. Using the selected sample, covariates are pulled from the VHA and CMS medical records then assigned in reference to the veteran’s first study month in the episode (e.g., veteran age, heart disease in past 12 months) and put into an analytic dataset.

Quality Control Process

Westat followed committee defined protocol to define study specific sample populations and conduct analyses. The lead researcher drafted initial specifications for the programmers to implement based on committee decisions. Iterations of the study protocol were implemented and made with committee consent. Upon completing analysis, Westat drafted results memos to share with the committee that included methodologic and programming notes where appropriate. The lead analyst and programmer wrote the memos, with one serving as the primary writer and the other in a quality control (QC) role. The results were also reviewed by the project director for quality control and privacy review processes.

SAS PROGRAMS:

An extensive number of scripts were written for this study and can be grouped by purpose. Tables H-1 and H-2 contain each script along with its name and purpose.

___________________

¹ U.S. Department of Health and Human Services. 2020. CMS Cell Suppression Policy. https://www.hhs.gov/guidance/document/cms-cell-suppression-policy (accessed September 19, 2024).

² National Center for Health Statistics. 2023. Underlying Cause of Death 1999-2020. https://wonder.cdc.gov/wonder/help/ucd.html#Source (accessed August 24, 2024).

1. Data Extraction and Transformation—veteran demographics and covariates were extracted from CDW and CMS, and they were saved to SAS datasets to later be used in assembling the analytic datasets. The team manually created ICD and Diagnosis reference files that were used by the scripts
   1. Demographic Variables
   2. Drug Variables
   3. Covariates defined by diagnostic codes
      1. Lookup tables
2. Identifying Eligible Veterans
   1. Inclusion/Exclusion
3. Creating Study 1 Analysis Data Sets
   1. Sample Frame
   2. Random sampling
   3. Creation of Analytic Files—Linkage of VHA, NDI and CMS files
   4. Fixes to code—additional derived variables
   5. Main analysis
   6. Sensitivity Analysis
4. Creating Study 2 Analysis Data Sets
   1. Sample Frame
   2. Sampling - First eligible episode
   3. Creation of Analytic Files—Linkage of VHA, NDI and CMS files
   4. Main analysis
   5. Bootstrap for main analysis to produce confidence intervals
   6. Sensitivity Analysis
5. Creating Study 3 Analysis Data Sets
   1. Sample Frame
   2. Sampling
   3. Creation of Analytic Files—Linkage of VHA, NDI and CMS files
   4. Main analysis
   5. Sensitivity Analysis

DATA PREPARATION PROGRAMS³

Table H-1 Data Preparation Programs

___________________

³ Note: Study 1 is referred to as trial 1, 4, and 5. Study 2 is referred to as trial 2. Study 3 is referred to as trial 3.

DATA ANALYSIS PROGRAMS⁴

Table H-2 Data Analysis Programs

___________________

⁴ Note: Study 1 is referred to as trial 1, 4, and 5. Study 2 is referred to as trial 2. Study 3 is referred to as trial 3.

This page intentionally left blank.