4098 — Dynamic causal inference for modeling complex management of patient care treatment using marginal structural models
Lead/Presenter: Tony Van,
COIN - Ann Arbor
All Authors: Van T (Center for Clinical Management Research, Ann Arbor VA), Wallace, B (Center for Clinical Management Research, Ann Arbor VA), Cohen-Mekelburg, S (Center for Clinical Management Research, Ann Arbor VA) Lipson, R (Center for Clinical Management Research, Ann Arbor VA) Wiitala, W (Center for Clinical Management Research, Ann Arbor VA) Waljee, A (Center for Clinical Management Research, Ann Arbor VA)
Veterans with inflammatory bowel disease (IBD) frequently have risk factors predisposing to both corticosteroid and anti-tumor necrosis factor (anti-TNF) side effects, including older age and comorbidities. Veterans are particularly vulnerable to long-term corticosteroid treatment. Prior work using Medicare and Medicaid data suggests that anti-TNF exposure reduces mortality in patients with Crohn's disease (CD) relative to prolonged corticosteroid exposure. We sought to reproduce this finding in a cohort of Veterans with IBD.
We developed a longitudinal cohort of 2,997 IBD patients between 2006 through 2015, comprised of 1,734 Crohn's disease (CD) and 1,263 ulcerative colitis (UC) patients. In our cohort, 1,161 patients initiated anti-TNF at the start of follow-up compared to 1,836 patients who entered as prolonged corticosteroids users. Among the corticosteroids users, 445 eventually switched to anti-TNF during follow-up. We estimated the effect of anti-TNF initiation on mortality using marginal structural modeling to control for 54 time-varying confounders which might affect treatment complications and a clinician's recommendations for therapy during follow-up. Baseline and time-varying propensity models for anti-TNF initiation and censoring were used to construct inverse probability of treatment weights for adjustment in the outcome analysis.
For CD, the estimated mortality hazard ratio for initiating anti-tnf relative to remaining on steroids was 0.57 (95% CI:.35, .94), p < 0.05. The estimated mortality hazard ratio for initiating anti-tnf in the UC cohort relative to remaining on steroids was 0.28 (95% CI: .10, .76), p < 0.05.
Anti-TNF exposure was associated with significantly reduced all-cause mortality rates compared with prolonged corticosteroid exposure. This is consistent with prior work using a civilian IBD cohort, though the effect for UC was not statistically significant in that population. Marginal structural modeling offers a unique way to address time-varying confounders and changing treatment exposure over time.
The survival benefit of anti-TNF therapy, as compared to the harms of corticosteroids is important to differentiate in the care of Veterans with IBD, as clinicians often rely on steroids given concerns regarding the safety of anti-TNF therapy in this vulnerable group.