Trends in Body Mass Index Before and After Diagnosis of Hidradenitis Suppurativa

S. Wright; A. Strunk; A. Garg

Disclosures

The British Journal of Dermatology. 2021;185(1):74-79. 

In This Article

Materials and Methods

This was a retrospective case–control study using a multiple healthcare system data analytics and research platform (Explorys) developed by IBM Watson Health (Armonk, NY, USA).[19] Clinical information from electronic medical records, laboratories, practice management systems, and claims systems was matched using the single set of Unified Medical Language System ontologies to create longitudinal records for unique patients. Data were standardized and curated according to common controlled vocabularies and classifications systems, including International Classification of Diseases (ICD), Systemized Nomenclature of Medicine–Clinical Terms,[20] Logical Observation Identifiers Names and Codes (LOINC)[21] and RxNorm.[22] The database encompasses more than 64 million unique lives, representing approximately 15% of the population across all census regions of the USA. Patients with all types of insurance in addition to those who fall into the self-pay category are captured.

The source population was composed of a 100% sample of patients with HS and a 10% sample of patients without HS from the Explorys database. The study population was limited to patients aged 18 years or older with active status in the database between 1 January 1999 and 9 September 2019. Patients with HS were identified using at least one code from the ICD, Ninth Revision (ICD-9; 705·83) or from the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10; L73·2). In an independent validation study of a separate electronic health record database, we observed a positive predictive value of 79·3% and an accuracy of 90% for diagnosis of HS using this algorithm.[23] The control cohort included patients with active status in the database who did not have a diagnosis code for HS. BMI measurements recorded during the course of care were identified using LOINC code 39156–5. To be eligible for the study sample, patients with HS must have had at least one BMI measurement for seven consecutive years before HS diagnosis, and three consecutive years after diagnosis. A longer period prior to diagnosis was chosen to account for potential delays in diagnosis.[2] Controls were required to have at least 10 consecutive years with available BMI data. 'Baseline' BMI was defined as the BMI measurement 7 years prior to diagnosis for patients with HS, and the first of the 10 consecutive BMI measurements for controls. We excluded patients with missing data on age, sex, race/ethnicity or diagnosis date. Eligible patients with HS and controls were matched at a ratio of 1 : 1 for age at baseline, sex, race and year of baseline BMI measurement within 2 years.

For all patients, one BMI measurement was identified each year for the 7 years before diagnosis and for the 3 years after diagnosis. If there were multiple BMI measurements within the same 1-year period, the BMI measurement closest to the period midpoint was selected. For controls, time intervals for BMI assessment were anchored around the patient's first BMI measurement, with the first BMI measurement serving as the midpoint of the patient's first assessment interval.

Statistical Analysis

BMI trends in patients with HS and controls were compared using separate linear mixed effects models (LMMs) for the periods before and after diagnosis. Random effects terms were included for participant identification and matched pair in order to account for multiple observations of the same participant and also for the correlation of participants within the same matched pair. BMI was the outcome variable in all models, and predictor variables included time (from baseline), HS diagnosis, age, sex, race, smoking status and time*HS interaction. Smoking was treated as a time-varying covariate, with patients classified as smokers after their first relevant diagnosis code. The LMMs were used to compare baseline BMI between patients with HS and controls, to assess whether BMI changed significantly during the period before or after diagnosis, and to compare the rate of BMI change between patients with HS and controls in each period. In a secondary subanalysis, we assessed whether differences in baseline BMI or rate of BMI change in patients with HS prior to diagnosis and controls depended on sex, race, age at diagnosis or baseline smoking status. In the demographic subanalyses, we do not report P-values for differences between patients with HS and controls within individual subgroups, as such tests are often constrained by limited power owing to smaller sample sizes, and can magnify the problem of multiple comparisons. Rather, we report interaction P-values, which assess whether the relationship between HS and BMI/change in BMI is the same for members of different demographic subgroups.

We additionally performed an ad hoc sensitivity analysis using an alternative definition of smoking status, in which patients were classified as smokers at all timepoints if they ever had a smoking diagnosis in their medical record. We also conducted a post hoc analysis to estimate difference in rate of BMI change between patients with HS and controls during the 3-year period after diagnosis, stratified by age at diagnosis. This exploratory analysis was performed after graphical inspection showed differences in BMI trends following HS diagnosis according to age. Accordingly, exploratory analysis did not include formal hypothesis tests. This study was approved by the human subjects committee at the Feinstein Institutes of Medical Research at Northwell Health.

processing....