Prior evidence on the association of analgesic medications and survival in EOC is from populations of predominantly White women. Black women with EOC experience lower survival rates compared to White women [7]. Yet, Black women are often underrepresented in ovarian cancer research, and to date, no studies have examined the role of analgesic medications in outcomes of Black women diagnosed with EOC. The present study investigates the relationship between pre-diagnostic, non-prescription analgesic medication use and survival among Black women with EOC.
We utilised data from the African American Cancer Epidemiology Study (AACES) [21], a population-based study of Black or African American women diagnosed with EOC. Study participants were identified through rapid case ascertainment at SEER or state cancer registries, gynaecologic oncology departments and hospitals. Eligible participants self-identified as Black or African American, were aged 20-79 years, were diagnosed with invasive EOC between December 2010 and December 2015, and resided in 11 geographic regions across the U.S.: Alabama, Georgia, Illinois, Louisiana, Michigan [Detroit], New Jersey, North Carolina, Ohio, South Carolina, Tennessee and Texas. A baseline telephone interview was completed to obtain information on demographics, reproductive history, medication use and lifestyle activities (e.g. smoking, physical activity). A shortened questionnaire was available for women who might otherwise have refused. Institutional Review Board approval was obtained at each site, and informed consent was obtained from all participants.
Participants were asked to recall any pain or inflammation medications they had taken regularly (at least once a week or at least 5 days per month) at any point in their lives. To aid in recollection, examples of these medications and their uses were provided. Participants who reported ever using such medications were asked to report the medication name, reason for use, age at first and last use and frequency and duration of use. This process was repeated for each medication used. The medications were then categorised into three groups: aspirin, naNSAIDs and acetaminophen. Combination medications (e.g. Excedrin includes aspirin and acetaminophen) were categorised as both medication types. To determine pre-diagnostic medication use, we categorised women who initiated analgesic medication use within the year before diagnosis or any time after diagnosis as non-users. We also considered women who reported a duration of use less than 6 months as non-users in order to capture the regular users of these medications. Since the short version of the questionnaire did not enquire about analgesic medication use, we only included participants who completed the long version of the questionnaire. Post-diagnostic analgesic medication use was not systematically collected in AACES and was not examined in this analysis.
Participant characteristics were summarised using descriptive statistics and compared by analgesic medication use using the Wilcoxon rank sum or chi-square tests. Vital status was determined using data from the National Death Index, cancer registries, the LexisNexis database and patient contact. Follow-up time was calculated as the time from the interview to death or the date of last contact. Multivariable Cox proportional hazard (PH) models were used to estimate hazard ratios (HR) and 95% confidence intervals (CI) for the association of analgesic medication use (ever, frequency [<30 and ≥30 days/month], and duration [≤5 years, >5 years]) with risk of all-cause mortality. The categories for frequency and duration of use were determined by prior literature and data distribution. Several potential confounders were considered in multivariable models, including age (continuous, years), geographic region (North [Illinois, Michigan, New Jersey and Ohio], Southeast [Georgia, North Carolina, South Carolina and Tennessee], and Southwest [Alabama, Texas, Louisiana]), stage (I, II, III and IV), histotype (high-grade serous carcinoma [HGSC]/carcinosarcoma, not HGSC/carcinosarcoma), smoking status (ever, never), physical activity (met vs. did not meet Physical Activity Guidelines for Americans [22]), body mass index ([BMI]; continuous, kg/m), Charlson comorbidity index [23] ([CCI]; 0, ≥1), education (high school graduate/GED or less, some college or college graduate), income (<$25,000, $25,000 to $74,999, ≥$75,000), insurance (any private insurance, any Medicaid and any Medicare as three separate variables) and debulking status (optimal, suboptimal). Of these, smoking status, physical activity, education and income did not appreciably change the results and were not included in the final models. Additionally, the use of each analgesic medication was included in the models to determine the independent effect of each analgesic medication, adjusting for the use of the others. Adjusted Kaplan-Meier curves were generated using the ggadjustedcurves() function in the survminer R package. The PH assumption was tested by evaluating Schoenfeld residuals and time x covariate interactions individually and collectively across all covariates. Histotype violated the PH assumption (p < 0.05) and was included as a strata term in the models. Stratified analyses were performed by histotype and the CCI (0 vs. ≥1).
Due to missing data on key covariates included in the models (ranging from 0.3% for insurance and 34% for debulking status), we imputed missing data using the multiple imputation by chained equations ('mice') R package. A Cox PH model was estimated, including the analgesic medications and all potential confounders described above, as well as the Nelson-Aalen estimate of the cumulative hazard [24]. The fraction of missing information was used to set the number of imputations (n = 39). The imputed and observed distributions of variables with missing data were comparable (Supplementary Table 1). As a sensitivity analysis, the primary analyses were repeated, restricting to participants with complete data.