Skip to main content
. Author manuscript; available in PMC: 2018 Dec 5.
Published in final edited form as: Ann Intern Med. 2018 May 8;168(11):783–790. doi: 10.7326/M17-3074

Appendix Table 2.

ROB Assessment in Studies That Reported on the Association Between PDMPs and Nonfatal and Fatal Drug Overdoses

Criteria for ROB
Assessment in Studies
That Reported on
PDMP Effects
Study, Year (Reference)
Maughan, 2015 (31) Bachhuber, 2016 (30) Brown, 2017 (23) Pauly, 2018 (28)

Bias due to confounding Serious – GEE model adjusted for calendar quarter, metropolitan area, interaction between calendar quarter and metropolitan area, area unemployment rate. Inadequate adjustment for co-implemented policies Serious – GEE model adjusted for calendar quarter, metropolitan area, interaction between calendar quarter and metropolitan area, area unemployment rate. Inadequate adjustment for co-implemented policies Critical – Failure to account for competing interventions that might have affected the rate of nonfatal and fatal opioid-related overdose events is likely to differentially affect the pre-intervention or post-intervention rate of events. Focus on slopes, instead of intercept, increases risk for time-varying confounding. Serious – GEE model adjusted for time, geographic region, rate of diagnosed substance use disorder, percentage of population male, percentage aged 25–35 years, and insured population counts. Inadequate adjustment for time-invariant factors and co-implemented policies.

Bias in selection of participants into the study Low – Selection of metropolitan areas based on data availability Low – Selection of metropolitan areas based on data availability Low – Single state examined pre-/post-intervention Moderate – Data from Truven Health MarketScan administrative claims data. Sample is representative of the privately insured and employed U.S. population

Bias in classification of interventions Low – Intervention was clearly defined. Low – Intervention was clearly defined. Low – Intervention was clearly defined. Low – Data from NAMSDL and PDAPS

Bias due to deviations from intended interventions Low – No deviation from the intended intervention. Low – No deviation from the intended intervention. Low – No deviation from the intended intervention. Low – No deviation from the intended intervention.

Bias due to missing data Low – No missing data were reported. Low – No missing data were reported. Low – No missing data were reported. Low – No missing data were reported.

Bias in measurement of outcomes Low – Measurement of outcome independent of policy Low – Measurement of outcome independent of policy Low – Measurement of outcome independent of policy Low – Measurement of outcome independent of policy

Bias in selection of the reported results Low – Expected analyses were reported. Low – Expected analyses were reported. Moderate – Post-implementation change in intercept was not reported. Low – Expected analyses were reported.

Overall bias Serious – Inadequate adjustment of time-invariant and time-varying factor Serious – Inadequate adjustment of time-invariant and time-varying factor Critical – Inadequate adjustment for key competing interventions. No alternative specifications were explored, e.g., date of legislation vs. implementation. Serious – Inadequate adjustment of time-invariant and time-varying factors; sample unlikely to represent U.S. population.
Paulozzi, 2011 (34) Kim, 2013 (19) Li, 2014 (35) Delcher, 2015 (29) and Delcher, 2016 (36)

Bias due to confounding Moderate – State fixed-effects adjust for time-invariant differences and adjustment for time, state, and spatial autocorrelation; however, inadequate adjustment for co-implemented policies Moderate – State fixed-effects adjust for time-invariant differences, adjustment for demographic, socioeconomic status, health/health care, and gun control laws; however, limited adjustment for time-varying factors Critical – GEE model adjusted for calendar quarter, demographic characteristics (percentage male, percentage aged 35–54 years, percentage white), geographic region, unemployment rate. Inadequate adjustment for time-invariant state differences and co-implemented policies Serious – Misspecification of ARIMA model poses serious risk; however, numerous sensitivity analyses were performed; minimal adjustment for co-implemented policies except pill mill laws.

Bias in selection of participants into the study Low – Selection of 50 U.S. states and D.C. Low – Selection of 50 U.S. states and D.C. Low – Selection of 50 U.S. states and D.C. Moderate – Study limited to state of Florida, unlikely to represent U.S. effect

Bias in classification of interventions Low – Intervention determined from IJIS Institute. Low – Intervention determined from IJIS Institute and NAMSDL Low – Intervention from DEA Low – Implementation date ascertained from state

Bias due to deviations from intended interventions Low – No deviation from the intended intervention Low – No deviation from the intended intervention Low – No deviation from the intended intervention Low – No deviation from the intended intervention

Bias due to missing data Low – No missing data were reported. Low – No missing data were reported. Low – No missing data were reported. Low – Minimal missing data

Bias in measurement of outcomes Low – Measurement of outcome independent of policy Low – Measurement of outcome independent of policy Low – Measurement of outcome independent of policy Low – Measurement of outcome independent of policy

Bias in selection of the reported results Low – Expected analyses were reported. Low – Expected analyses were reported. Low – Expected analyses were reported. Low – Expected analyses were reported.

Overall bias Moderate – Inadequate adjustment of time-varying factor, but assessment of spatial autocorrelation Moderate – Inadequate adjustment of time-varying factor Serious – Inadequate adjustment of time-invariant and time-varying factors Serious – Misspecification of ARIMA model will produce bias estimates; sensitivity analyses reduce some ROB.
Radakrishnan, 2015 (21) Kilby, 2015 (32) Patrick, 2016 (27) Birk, 2017 (20)

Bias due to confounding Low – State and year fixed-effects adjust for time-invariant differences; robust adjustment for sociodemographics and adjustment for co-implemented policies Moderate – State and fixed-effects adjust for time-invariant differences, adjustment for unemployment rate and population over 60 years; limited adjustment for time-varying factors Moderate – State fixed-effects adjust for time-invariant differences, adjusted for unemployment, education attainment rate; extensive sensitivity analyses Low – State fixed-effects adjust for time-invariant differences; adjustment for pill mill laws, Good Samaritan laws, naloxone distribution, and medical marijuana legalization

Bias in selection of participants into the study Low – Selection of 50 U.S. states and D.C. Moderate – 12 states that adopted PDMPs before 2003 were excluded. Low – 15 states that did not implement a law during study period were excluded. Low – Selection of 50 U.S. states and D.C.

Bias in classification of interventions Low – Intervention determined from NAMSDL Low – Intervention determined from NAMSDL Low – Intervention from Law Atlas and NAMSDL Low – Intervention determined from The Network of Public Health Laws and the PDMP Center for Excellence

Bias due to deviations from intended interventions Low – No deviation from the intended intervention Low – No deviation from the intended intervention Low – No deviation from the intended intervention Low – No deviation from the intended intervention

Bias due to missing data Low – No missing data were reported. Moderate – 12 states that adopted PDMP before 2003 Moderate – Excluded Florida and West Virginia due to influence of co-implemented laws; some suppressed data, multiple imputation Low – No missing data were reported.

Bias in measurement of outcomes Low – Measurement of outcome independent of policy Low – Measurement of outcome independent of policy Low – Measurement of outcome independent of policy Low – Measurement of outcome independent of policy

Bias in selection of the reported results Low – Expected analyses were reported. Low – Expected analyses were reported. Low – Expected analyses were reported. Low – Expected analyses were reported.

Overall bias Low – Adequate adjustment for time-invariant factors (fixed effects) and robust control for co-implemented policies Moderate – Inadequate adjustment of time-varying factor; exclusion of 12 early-adopter states Moderate – Inadequate adjustment of time-varying factors; exclusion of states with time-varying confounding instead of adjustment Low – Robust adjustment of confounding and sensitivity analyses
Nam, 2017 (33) Meinhofer, 2017 (22) Dowell, 2016 (25) Pardo, 2017 (26)

Bias due to confounding Moderate – State and year fixed-effects adjust for time-invariant differences and secular trends; adjustment for sociodemographics and health care access; however, inadequate adjustment for co-implemented policies Moderate – State and year fixed-effects adjust for time-invariant differences and secular trends; however, adjustment for co-implemented policies Moderate – State and year fixed-effects adjust for time-invariant differences and secular trends, adjustment for opioid prescribing rate, pending death rates; however, inadequate adjustment for co-implemented policies Low – State and year fixed-effects adjust for time-invariant differences and secular trends; adjustment for pill mill laws, Good Samaritan laws, naloxone distribution, and medical marijuana legalization

Bias in selection of participants into the study Moderate – 15 states that adopted PDMPs before 2000 were excluded. Low – Selection of 50 U.S. states and D.C. Low – 12 states were excluded for low outcome counts and complex policy situations. Low – Selection of 50 U.S. states and D.C.

Bias in classification of interventions Moderate – No information on how date of intervention was determined Low – Intervention determined from NAMSDL, TTAC, and ONCHIT Low – Intervention from WestLaw, TTAC, and NAMSDL Critical – The continuous score that was the primary independent variable of interest was determined on the basis of factors previously found to reduce the outcome.

Bias due to deviations from intended interventions Low – No deviation from the intended intervention Low – No deviation from the intended intervention Low – No deviation from the intended intervention Low – No deviation from the intended intervention

Bias due to missing data Moderate – Some data suppressed due to small counts; 15 states that adopted PDMPs before 2000 were excluded. Moderate – States with zero counts were imputed 1 dead. Moderate – 12 states with low number of deaths were excluded. Low – No missing data were reported.

Bias in measurement of outcomes Low – Measurement of outcome independent of policy Low – Measurement of outcome independent of policy Low – Measurement of outcome independent of policy Low – Measurement of outcome independent of policy

Bias in selection of the reported results Low – Expected analyses were reported. Low – Expected analyses were reported. Low – Expected analyses were reported. Serious – The use of a subjective exposure increases the likelihood that many analyses were conducted to determine the best model; however, no information on such analyses was reported.

Overall bias Moderate – Inadequate adjustment for time-varying factors, exclusion of states that adopted PDMP before 2000 Moderate – Inadequate adjustment of time-varying factor; unexpected handling of missing data Serious – Inadequate adjustment of time-varying factor; unexpected handling of missing data Serious – Subjective measure of intervention; inadequate information on multiple tests; robust adjustment of confounding
Phillips, 2017 (24)

Bias due to confounding Critical – Random state and year variable fail to account for time invariant differences among states or secular trends; adjustment for education, unemployment, percentage of population on disability, and medical marijuana laws; however, no adjustment for other time-varying policies

Bias in selection of participants into the study Low – Selection of 50 U.S. states and D.C.

Bias in classification of interventions Low – Intervention determined from NAMSDL

Bias due to deviations from intended interventions Low – No deviation from the intended intervention

Bias due to missing data Low – No missing data were reported.

Bias in measurement of outcomes Low – Measurement of outcome independent of policy

Bias in selection of the reported results Low – Expected analyses were reported.

Overall bias Critical – Inadequate adjustment of time-invariant and time-varying factor

ARIMA = autoregressive integrated moving average; DEA = U.S. Drug Enforcement Administration; GEE = generalized estimating equation; IJIS = Integrated Justice Information Systems; NAMSDL = National Alliance for Model State Drug Laws; ONCHIT = Office of the National Coordinator for Health Information Technology; PDAPS = Prescription Drug Abuse Policy System; PDMP = prescription drug monitoring program; ROB = risk of bias; TTAC = Brandeis University's PDMP Training and Technical Assistance Center.