Data source and study design
We examined specific LD-associated clinical characteristics using the Clinformatics® Data Mart database from January 1, 2018, to December 31, 2019. Clinformatics Data Mart is an integrated US healthcare claims database that includes privately insured enrollees with commercial or Medicare Advantage plans affiliated with Optum. The database contains statistically de-identified and Health Insurance Portability and Accountability Act-compliant medical claims, including healthcare services performed in inpatient and outpatient settings, and pharmacy claims for approximately 15–20 million individuals annually. The Clinformatics Data Mart population is geographically representative, spanning all 50 US states. In addition to medical and pharmacy claims, the database includes provider data and information on patient enrollment and demographics [14]. Clinformatics Data Mart utilizes internal and external sources to comprehensively capture death information, such as the Death Master File maintained by the Social Security Office, enrollment data files in Medicare data, obituary data, facility claims, member coverage information, and Optum EHR data.
Study population
First, we assembled an all-adult cohort, including all eligible adult individuals (aged ≥ 18 years on January 1, 2018) who continuously enrolled from January 1, 2018 until December 31, 2019 (or until they died, if this was before the end date of the study period), allowing a 45-day enrollment gap (Fig. 1). Next, a cohort comprising patients with a diagnosis of HIV disease was identified to include individuals who had at least 1 inpatient or 2 outpatient HIV diagnoses (i.e. International Classification of Diseases [ICD]-9-CM: 042, 079.53, V08; ICD-10-CM: B20, B97.35, Z21) on separate calendar dates among the all-adult cohort. This identification algorithm was modified from the validated Medicaid-based algorithm to identify people living with HIV [15, 16]. An LD cohort was also created from the all-adult cohort by identifying LD diagnosis (i.e. ICD-9-CM: 272.6; ICD-10-CM: E88.1) in medical claims, or metreleptin prescription (National Drug Code: 66780-310-01, 76431-210-01) in pharmacy claims. In order to increase specificity and decrease misclassification due to rule-out diagnosis, at least 1 inpatient or 2 outpatient LD diagnoses on separate calendar dates were required. The LD cohort was further divided into non-HIV-LD and HIV-LD subgroups. We defined the non-HIV-LD group as individuals without any HIV diagnosis, and the HIV-LD group as individuals with at least 1 inpatient or 2 outpatient HIV diagnoses among the LD cohort. By definition, the non-HIV-LD group included all individuals with GLD or non-HIV-associated PLD, and the HIV-LD group included HIV-associated lipoatrophy and lipohypertrophy.
Flow chart for clinical characteristic analysis in 2018–2019. aIndividuals who had at least 1 inpatient or 2 outpatient HIV diagnoses (i.e. ICD-9-CM: 042, 079.53, V08; ICD-10-CM: B20, B97.35, Z21) on separate calendar dates are qualified to have HIV. Abbreviations: HIV-LD, lipodystrophy associated with HIV; ICD, International Classification of Diseases; non-HIV-LD, lipodystrophy not associated with HIV; PWHD, people living with HIV disease
Individuals in the non-HIV-LD group were matched by age and sex in a 1:4 ratio to controls from the all-adult cohort without HIV. The controls were required to have neither an LD nor an HIV diagnosis. To minimize the effect of fat alterations associated with HIV infection and antiretroviral therapy, individuals in the HIV-LD group were matched by age and sex in a 1:4 ratio to controls from the people living with HIV cohort. The people living with HIV controls were required not to have an LD diagnosis.
Descriptive data analysis 2018–2019
We examined patient demographics (i.e. age, sex, and race/ethnicity), estimated their Elixhauser Comorbidity Index [17] (identification of 38 different pre-existing conditions based on secondary diagnoses) and estimated the prevalence of clinical characteristics among the non-HIV-LD and HIV-LD cohorts. The specific clinical characteristics evaluated included hyperlipidemia, diabetes mellitus, hypertension, acute myocardial infarction (AMI), MASLD or metabolic dysfunction-associated steatohepatitis (MASH), liver fibrosis or cirrhosis, acute pancreatitis, kidney disease (i.e. acute or chronic glomerulonephritis, acute or chronic renal failure, nephritis or nephrotic syndrome, renal failure or proteinuria), autoimmune diseases (i.e. systematic lupus erythematosus [SLE], rheumatoid arthritis, autoimmune thyroiditis), cancers excluding non-melanoma skin cancers, and serious infections (i.e. bacteremia, pneumonia, skin/soft tissue infection, gastrointestinal infection, acute osteomyelitis, acute pyelonephritis, acute meningitis) resulting in hospitalization. We examined the prevalence of overall cancer and specific types of cancers (i.e. lymphoma, breast cancer, prostate cancer). Lymphoma was identified with at least 1 inpatient diagnosis or 2 outpatient lymphoma diagnoses on separate calendar dates occurring at least 2 months apart. This approach was a modified version of a validated algorithm developed by Setoguchi et al., expecting > 80% sensitivity and > 90% specificity [18]. We also applied the same identification algorithm to other cancers [18]. To identify AMI, we applied an identification algorithm yielding 86.0% positive predictive value from a primary hospital discharge diagnosis of AMI [19]. Acute pancreatitis was identified from a primary hospital discharge diagnosis of acute pancreatitis [20]. To identify SLE, we modified a claims-based identification algorithm by using at least 1 inpatient diagnosis or 2 outpatient diagnoses at least 2 months apart [21].
Furthermore, we used a case definition of serious infections resulting in hospitalizations [22]. For hyperlipidemia, diabetes mellitus, and hypertension, we used the Centers for Medicare and Medicaid Services (CMS) Chronic Condition Warehouse (CCW) algorithms using at least 1 inpatient or 2 outpatient diagnoses on separate calendar dates [23].
Statistical analysis
The patient demographics and clinical characteristics among non-HIV-LD and HIV-LD cases were presented as percentages for categorical variables, and as means and standard deviations for continuous variables. We compared demographics and clinical characteristics between the LD cohorts and matched controls using the Chi-square test, Fisher’s exact test, and student t-test, as appropriate. Logistic regression models were used to estimate the odds ratios (ORs) and accompanying 95% confidence intervals (CIs) for each clinical characteristic among non-HIV-LD and HIV-LD cohorts compared with the matched controls. All analyses were performed using Health Data software (Panalgo, Boston, MA, USA) with a two-sided level of statistical significance of 0.05.
Data and resource availability
All data analyzed during this study are from the Clinformatics Data Mart database that is cited in the Methods section. All analyses conducted in this study are presented in the manuscript.
Add Comment