Exploring The Association Between Antidepressants and Colorectal Cancer in Administrative Data: Negative Controls, Active Comparators and Algorithms
Public Deposited- Last Modified
- March 20, 2019
- Creator
-
D'Arcy, Monica
- Affiliation: Gillings School of Global Public Health, Department of Epidemiology
- Abstract
- Some antidepressants, especially Selective Serotonin Reuptake Inhibitors (SSRIs), may prevent colorectal cancer (CRC), but these effects may be drug rather than class specific. Previous epidemiological studies have only examined class-level effects, and all studies used non-user comparisons, which are susceptible to several biases. Examining specific SSRI-CRC associations requires a large sample size and precise prescription records, which are features of administrative data; however, these data do not generally contain pathology confirmed cases and algorithms are required to identify probable cases. The goals of this dissertation were: 1) to examine the class-level associations between three antidepressant classes, including SSRIs, and CRC compared to a negative control, antihypertensive initiators (AHT), 2) to examine the association between specific SSRIs and CRC, and 3) to re-evaluate claims-based CRC-identification algorithms in a contemporary population. To examine the first two goals, we performed a new-user, cohort study using a 20% random sample of Medicare beneficiaries (2007-2013), aged ≥66. We estimated hazard ratios (HRs) and 95% confidence intervals (CI), and controlled measured confounding using a propensity score weighting approach. SSRI initiators had lower CRC rates compared with AHT initiators (aHR=0.85, 95% CI: 0.70-1.02). Paroxetine and fluoxetine initiators had lower CRC rates compared with citalopram users (aHR: 0.78, 95% CI: 0.56-1.06; aHR: 0.74, 95% CI: 0.52-1.05, respectively). Estimates were consistent across sensitivity analyses. We re-evaluated CRC-identification algorithm performance in a ≥65, 2006-2009 North Carolina Medicare population, a proportion of which were cancer registry identified CRC cases. We employed a novel cohort creation strategy, whereby cases contribute information from both their pre-diagnostic non-case and case states to accurately capture CRC incidence. Specificity was lower (98.3-99.4% versus 98.5-99.6%) and Positive Predictive Value (PPV) substantially lower (18-37% versus 45-71%) in this population compared to the original population. Results from the first two goals warrant further investigation into the SSRI-CRC association, including incorporating additional part D data as it becomes available. Algorithms are a necessity when performing a drug-cancer study in administrative data, but should be used cautiously, because they are population and time specific. These CRC-identification algorithms need to be updated to reflect a more contemporary and economically diverse population. Future validation studies should employ strategies to accurately ascertain incidence to avoid overestimating PPV.
- Date of publication
- December 2016
- Keyword
- DOI
- Resource type
- Rights statement
- In Copyright
- Advisor
- Sandler, Robert
- Troester, Melissa
- Meyer, Anne Marie
- Baron, John
- Lund, Jennifer
- Stürmer, Til
- Degree
- Doctor of Philosophy
- Degree granting institution
- University of North Carolina at Chapel Hill Graduate School
- Graduation year
- 2016
- Language
Relations
- Parents:
This work has no parents.
Items
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
DArcy_unc_0153D_16646.pdf | 2019-04-12 | Public | Download |