2012 National Meeting

3038 — Method of Analysis as a Factor in Models of Admission among Veterans of Iraq and Afghanistan Deployments

Stock EM, Central Texas Veterans Healthcare System and Scott & White Healthcare, Temple, TX; Cooper TL, Carl R. Darnall Army Medical Center, Fort Hood, TX; Tsan JY, VISN 17 VA Center of Excellence for Research on Returning War Veterans, Waco, TX; Zeber JE, and Copeland LA, Central Texas Veterans Healthcare System and Scott & White Healthcare, Temple, TX;

Objectives:
Pharmacological treatment for post-traumatic stress disorder (PTSD) is complicated by many treatment options with limited efficacy. It is uncertain, however, how the use of antidepressant monotherapy and concurrent psychotropics is associated with hospital admissions that may result from drug-associated adverse events, treatment non-response, or other causes. To illuminate the impact of pharmacological treatment for PTSD, we modeled all-cause admission for a cohort of OEF/OIF Veterans using several methodological approaches.

Methods:
We obtained data on 398 OEF/OIF Veterans diagnosed with PTSD in the South Texas VA during fiscal year 2006 (Oct 2005-Sep 2006). Models controlled for patient demographics and comorbidities, contrasting high-dimensional propensity score (hd-PS) approaches (standardized mobility ratio (SMbR)-weighted, inverse-probability-of-treatment-weighted (IPT), unweighted) with traditional covariate-adjusted models. Treatment measures included first-line antidepressant monotherapy per VA/DoD Clinicial Practice Guidelines (CPG) and first-line polypharmacy.

Results:
Traditional models explained less variance than hd-PS models, suggesting a strong influence of methodology on the interpretation of the multivariate results. In the unweighted hd-PS model, Veterans appeared to be at a greater risk of all-cause admission if concurrently prescribed psychotropics rather than first-line monotherapy (Odds Ratio [OR] = 2.76; 95% Confidence Interval [CI]: 1.08-7.01), but this effect was not observed with the traditional covariate-adjusted logistic model (OR = 2.12; 95% CI: 0.91-4.95). Both SMbR (OR = 5.26; 95% CI: 2.66-10.41) and IPT-weighted (OR = 4.04; 95% CI: 2.30-7.10) hd-PS models estimated an increased risk of admission associated with concurrent psychotropics.

Implications:
In this era of comparative effectiveness research, not only does study design wield influence over the apparent clinical meaning of research results but also analytic approach. The hd-PS approach demonstrated a remarkable ability to balance a multitude of patient characteristics and appeared advantageous for small samples with fewer outcomes, whereas conventional models fall short.

Impacts:
It is important to consider hd-PS analytic approaches in addition to traditional approaches, as we found potentially greater risks of admission among patients prescribed a variety of additional psychotropics despite CPG recommendations. To avoid adverse outcomes, one should continuously assess and reevaluate patients' medication history, especially regarding the use of concurrent psychotropics with first-line antidepressants.