California PDMP Enhancement, Analysis and Response Initiative California Department of Justice University of California Davis Harold Rogers PDMP National Meeting August 15-16, 2016
Outline 1. Project overview 2. Evaluating de-duplication of PDMP data 3. Integrating PDMP and hospital / emergency department data 4. Integrating data from other states
CURES 2.0 California SB 809 (stats 2013, Chapter 400, DeSaulnier) was signed by the Governor on September 27, 2013: 1. Re-funded PDMP operations 2. Mandatory provider registration by July 1, 2016 3. Enabled CA DOJ to add advanced features: Updated user interface Automatic (unsolicited) alerts within PDMP Ability to send peer-to-peer messages within PDMP
California Harold Rogers Grant Partnership between California DOJ and UC Davis researchers Evaluate implementation and effect of CURES 2.0 on prescribing practices drug overdose rates Optimize PDMP utility by coordinating with public health, regulatory, and law enforcement agencies Design and test data-driven algorithms for identifying high-risk prescribing and dispensing patterns
De-duplication of PDMP data Purpose: to identify prescriptions for the same individual Challenges No unique patient identifier Variation in identifying data for an individual Hundreds of millions of records
Example First Name Last Name Sex DOB Address Zip Code Stephen Henry Male 05/11/77 2450 48 th Street 95817 Steven Henry Male 05/11/77 2450 48 th St. 95817 Henry Stevens Male 11/05/77 2450 48 th St., Apt. 2 95817 Steve Henry Male 05/11/87 2405 48 th Street 95807 Are these the same person?
Strategy: Compare record linkage programs CURES 2.0 custom built program SAS application The Link King: http://www.the-link-king.com/index.html SAS application Link Plus: http://www.cdc.gov/cancer/npcr/tools/registryplus/lp.htm Microsoft Windows stand-alone application LinkSolv: http://www.strategicmatching.com/products.html Microsoft Access application
Determine for each program Blocking variables What matches to consider Cut-points for probabilistic matches How good does the match have to be Feasibility program run times, ease of use
Comparison procedure Stratify data by certainty of linkage From high to low probability of a match 4 reviewers will inspect a stratified random sample of matches Identity of the program and certainty of the match is withheld Truth determined by majority opinion
Statistics to compare Sensitivity: proportion of true matches identified by the program Specificity: proportion of true non-matches identified by the program Positive predictive value: proportion of identified matches that are true matches Negative predictive value: proportion of identified non-matches that are true nonmatches Area under ROC curve
Goals Determine which program is most accurate and feasible to use in production Inform best practices for PDMP record linkage
Integrating PDMP & outcome data
Analysis plan time series Track data from multiple sources PDMP registration & use (CURES) Prescribing patterns (CURES) Hospitalization & emergency department visits related to opioid-overdose (OSHPD) Overdose deaths (CDPH) Identify major events (e.g., mandatory PDMP registration) to estimate the effect on the indicators
Follow Counties Over Time Include county indicators into the time series. Explore each county s contribution to overlying trends.
Integrating data from other states Extend time series by including prescribing and overdose data from other states without the specific enhanced PDMP features of CURES 2.0 Compare trends in California counties (intervention group) with counties in other PBSS states (control group) Takes into account potential changes affecting both groups (e.g., national policy trends)
Analysis plan Estimate the effect CURES 2.0 enhancements on prescribing practices, fatal overdoses, and non-fatal overdose. Use data from PBSS, NSDUH, & CDC WONDER Use a difference-in-difference approach to isolate effects associated with CURES 2.0 enhancements.
Example 1: Time series analysis (single group) 0 5 10 15 20 25 30 Intervention starts Apparent effect of the intervention 1 3 5 7 9 11 13 15 17 19 21 Time Rate intervention group
0 5 10 15 20 25 30 Example 2: Time series analysis (with comparison group) Intervention starts No effect of the intervention. Something else could have happened at the same intervention period affecting both groups 1 3 5 7 9 11 13 15 17 19 21 Time Rate intervention group Rate control group
Questions California Department of Justice Mike Small (mike.small@doj.ca.gov) Tina Farales (tina.farales@doj.ca.gov) University of California Davis Stephen Henry (sghenry@ucdavis.edu) Alvaro Castillo (alvacasti@gmail.com)