Wednesday, May 15, 2013
Present: Marianne Burke, Abby Crocker, Kairn Kelley, Amanda Kennedy, Ben Littenberg, Charlie MacLean, Connie van Eeghen
1. Start Up: Connie reported back from a poster presentation in Denver; great opportunity to meet new colleagues
2. Presentation: Charlie MacLean: Exploration of analytical plan for Natural History of Acute Opioid Use
a. Key points from the 2009 Boudreau article that Amanda found in her lit review: Trends in De-facto Long-term Opioid Therapy for Chronic Non-Cancer Pain (CNCP)
i. CONSORT study of adults enrolled in two health plans serving > 1% of US population (4m)
ii. Automated health plan data to construct episodes of opioid use 1997-2005; incidence & prevalence; rates of change
iii. Long term episode: >90 days with 120+ days’ supply or 10+ opioid rx in one year
iv. Results: incident long term use increased from 8.5 to 12.1/1000 (Group Health) with 6% annual change (PCA) and 6.3 to 8.6/1000 (Kaiser Permanente) with 5.5 PCA
1. Small increases observed in overall use of opioids: 2.2 PCA and 0.83 PCA
2. Prevalent long term use almost doubled: 8.1 PCA and 8.6 PCA
3. Incident long term use lasted > 1 year for 75% of episodes
v. Non-Schedule II opioids most common and increased in long term use
1. MEq dose was stable for incident long term opioid users
vi. Regular users of sedative hypnotics: 28.6% (GH) and 30.2% (KPNC)
vii. Conclusion: increasing prevalence; need to study benefits and risks. Need to study concurrent use of opioids and sedative-hypnotics, which were unexpectedly common
b. Amanda provide a draft introduction and her search strategy.
i. Target population: adults getting their first dose of opioids with the intention of short term management.
c. Research Question: what proportion of opioid naïve adult patients started on an opioid become long term users? What are the characteristics of use? What are the predictors of long term use?
i. Previous study (Boudreau, above) provides definitions for “long term,” although we may not be able to determine “days’ supply.” Can be estimated and evaluated with a sensitivity analysis. We can look only at first to last claim, not scripts.
ii. Can be constructed as a survival analysis, with the “survivors” being those who continue to generate additional claims.
d. Ben & Abby sketched out statistical analyses
i. Kaplan-Meier survival analysis
ii. Logistic regression based on whether or not patient became long term/chronic, with variables of interest:
1. Should cancer-related opioid use be included? Literature is separate. We cannot easily determine patients considered terminally ill and long term consequences of opioid use for this population is not a great concern. May be identifiable by provider or treatments. Retain the data and examine: does the inclusion of cancer patients make a difference? Plan to stratify on cancer patients.
2. Eliminate non-opioid patients.
3. Eliminate non-opioid medications as inclusion criteria; keep as covariates. Include all opioids, including non-Schedule II.
4. Classify each episode as acute or chronic.
5. Collect predictive data: age, sex, nearness of provider (zip code), prior episodes, prior rx, prior chronic episodes, recent diagnoses (low back pain, headache), other med use, prescriber (specialty: PCP, ED, other; buprenorphine prescribers; methadone prescribers; chronicity index-preference for prescribing opioids), length of days of dose, dosage, insurer, co-morbid mental health conditions (anti-depressant use, diagnosis within past six months… but keep benzo’s separate), history of substance use disorder (previous SA claims), zip code linked to median income and education; out-of-state pharmacy (mail or picked up) based on pharmacy zip code
6. Drug index: was the initial drug a long or short acting formulation?
7. Hospital initiation: Will not see hospital prescriptions but will include hospital claims. 96 hours post-hospital filled rx is likely to be a hospital-initiated drug.
iii. Charlie: how do the prescribers cluster in prescribing patterns or preferences
e. Specification for data report
i. One record for each claim, linking medication data and patient data: Charlie and Abby over the next two weeks (on May 17th)
1. Steve Kappel this morning released a 5 gig data base for Charlie to review: all prescriptions from all prescribers; not the other claims: Charlie will import and clean
ii. Test data for face validity BEFORE excluding any claims from the study
1. Clean in STATA: Abby and Charlie
2. Run descriptive analyses – in CROW: Abby and Charlie
iii. Meet with Neil S on validation plan (Abby to join biweekly Tuesday meetings)
1. Request IRB permission for validation process
2. Validation process needed to confirm that patient identifiers are unique (not shared with other patients ) and not duplicated (only one per patient). May need to cross-validate with other data sources, under supervision by the IRB.
a. Use four years of data to build predication model; test model on data from year 5 – or hold out 20% of data, to avoid problems with a time trend – or both
b. One time related issue is the policy decision to avoid prescribing NSAIDs to elderly (over 65) in order to avoid stomach and kidney problems, resulting in greater use of opioids
iv. Analysis – CROW session TBD
3. Summer sessions: consider Wed 11:30 – 1:00 starting June 5. Will ask Sylvie for help in figuring this out.
a. May 16: Connie: Feedback on poster presentation on “Quick Turns in Tight Places: Implementing Change in Small Practices” and the Introduction and Methods sections of manuscript with working title: “Integrating Behavioral Health using Workflow (Lean)”
b. May 23:
c. May 30:
d. June 5: New summer schedule will start
e. Future agenda to consider:
i. Christina Cruz, 3rd year FM resident with questionnaire for mild serotonin withdrawal syndrome?
ii. Peter Callas or other faculty on multi-level modeling
iii. Charlie MacLean: demonstration of Tableau
iv. Journal article: Gomes, 2013, Opioid Dose and MVA in Canada (Charlie)
Posted by Connie at 5/15/2013 08:25:00 AM