Clinical guidelines for the use of opioids in chronic non-cancer pain recommend assessing risk for aberrant drug-related behaviors (ADRBs) prior to initiating opioid therapy. Despite recent drastic increases in prescription opioid misuse and abuse, the use of screening tools in assessing risk of ADRB by clinicians continues to be underutilized. We hypothesized that Natural Language Processing (NLP) techniques may support clinicians in risk assessment of patients considered for opioid therapy. Using a retrospective cohort of 3,672 chronic non-cancer pain patients with at least one Opioid Agreement (OA) between 1/1/2007 and 12/31/2012, we examined the availability of EHR structured and unstructured data to populate three ADRB risk tools, which varied in length, complexity and assessed patient life events: Opioid Risk Tool (ORT); Diagnosis, Intractability, Risk, Efficacy (DIRE); and Screener and Opioid Assessment in Patients with Pain (SOAPP). We developed structured data queries and NLP algorithms for processing unstructured data for each tool, and evaluated performance of these tools in predicting future ADRBs. Sufficient EHR structured and unstructured data existed to populate the ORT and DIRE, but not the SOAPP. At the time of their most recent OA, ORT-based classification identified 41.8% of patients as low risk, 28.2% moderate risk and 29.0% high-risk for ADRBs. DIRE classification identified 41.2% of patients as unsuitable (score <14) and 58.8% as possible (score 14+) candidates for long-term opioid therapy. During a year following the OA, 22.2% of patients had ADRBs. Compared to ORT low risk patients, moderate/high risk patients were 2.2/4.8 times more likely to have ADRBs, respectively. Patients with DIRE scores <14 were 2.9 times more likely to have ADRBs than those with scores 14+. Our findings suggest that NLP techniques have potential utility to support clinicians in screening chronic non-cancer pain patients considered for opioid therapy. Study supported by Pfizer Inc. Clinical guidelines for the use of opioids in chronic non-cancer pain recommend assessing risk for aberrant drug-related behaviors (ADRBs) prior to initiating opioid therapy. Despite recent drastic increases in prescription opioid misuse and abuse, the use of screening tools in assessing risk of ADRB by clinicians continues to be underutilized. We hypothesized that Natural Language Processing (NLP) techniques may support clinicians in risk assessment of patients considered for opioid therapy. Using a retrospective cohort of 3,672 chronic non-cancer pain patients with at least one Opioid Agreement (OA) between 1/1/2007 and 12/31/2012, we examined the availability of EHR structured and unstructured data to populate three ADRB risk tools, which varied in length, complexity and assessed patient life events: Opioid Risk Tool (ORT); Diagnosis, Intractability, Risk, Efficacy (DIRE); and Screener and Opioid Assessment in Patients with Pain (SOAPP). We developed structured data queries and NLP algorithms for processing unstructured data for each tool, and evaluated performance of these tools in predicting future ADRBs. Sufficient EHR structured and unstructured data existed to populate the ORT and DIRE, but not the SOAPP. At the time of their most recent OA, ORT-based classification identified 41.8% of patients as low risk, 28.2% moderate risk and 29.0% high-risk for ADRBs. DIRE classification identified 41.2% of patients as unsuitable (score <14) and 58.8% as possible (score 14+) candidates for long-term opioid therapy. During a year following the OA, 22.2% of patients had ADRBs. Compared to ORT low risk patients, moderate/high risk patients were 2.2/4.8 times more likely to have ADRBs, respectively. Patients with DIRE scores <14 were 2.9 times more likely to have ADRBs than those with scores 14+. Our findings suggest that NLP techniques have potential utility to support clinicians in screening chronic non-cancer pain patients considered for opioid therapy. Study supported by Pfizer Inc.