Prescription Drug Monitoring Programs and Death Rates from Drug Overdose


  • The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the Centers for Disease Control and Prevention.

  • No financial disclosures were reported by the authors of this article.

Leonard J. Paulozzi, MD, MPH, Division of Unintentional Injury Prevention, National Center for Injury Prevention and Control, Centers for Disease Control and Prevention, 601 Sunland Park Drive, Suite 200, El Paso, TX 79912, USA. Tel: 770-365-7616; Fax: 915-834-5973; E-mail:


Objective.  Drug overdoses resulting from the abuse of prescription opioid analgesics and other controlled substances have increased in number as the volume of such drugs prescribed in the United States has grown. State prescription drug monitoring programs (PDMPs) are designed to prevent the abuse of such drugs. This study quantifies the relation of PDMPs to rates of death from drug overdose and quantities of opioid drugs distributed at the state level.

Design.  Observational study of the United States during 1999–2005.

Outcome Measures.  Rates of drug overdose mortality, opioid overdose mortality, and opioid consumption by state.

Results.  PDMPs were not significantly associated with lower rates of drug overdose or opioid overdose mortality or lower rates of consumption of opioid drugs. PDMP states consumed significantly greater amounts of hydrocodone (Schedule III) and nonsignificantly lower amounts of Schedule II opioids. The increases in overdose mortality rates and use of prescription opioid drugs during 1999–2005 were significantly lower in three PDMP states (California, New York, and Texas) that required use of special prescription forms.

Conclusions.  While PDMPs are potentially an important tool to prevent the nonmedical use of prescribed controlled substances, their impact is not reflected in drug overdose mortality rates. Their effect on overall consumption of opioids appears to be minimal. PDMP managers need to develop and test ways to improve the use of their data to affect the problem of prescription drug overdoses.


Increases in prescription drug overdoses have driven a steep rise in the rate of drug overdose mortality in the United States in the past decade with much of the increase attributable to prescription opioid analgesics [1–4]. Nonsuicidal prescription opioid overdose deaths increased by 142% during the period 1999–2004, while heroin deaths declined [2]. The increasing numbers of opioid-related deaths were associated with parallel increases in both the prescribing of opioids [4] and the self-reported nonmedical use of these drugs [5]. Persons dying of prescription drug overdoses generally have a history of abusing or misusing the drugs and frequently obtaining them without prescriptions [6,7].

Prompted in part by the diversion of prescription opioids and other pharmaceuticals to nonmedical use, Congress asked the U.S. General Accounting Office (GAO) to study state prescription drug monitoring programs (PDMPs). The GAO concluded in 2002 that PDMPs were useful in reducing drug diversion [8]. State PDMPs have since proliferated in the United States, operating in 16 states in 2000 [9] and in 32 states by 2008 [10]. The Department of Justice instituted the Harold Rogers Prescription Drug Monitoring Program to help fund PDMPs in fiscal year 2003 [11], and the Department of Health and Human Services began to fund PDMPs through the National All Schedules Prescription Electronic Reporting Act (NASPER) in 2009 [12].

Although program specifics vary substantially from state to state, PDMPs typically require retail pharmacists to enter data from prescriptions for controlled substances into a centralized electronic database. These data identify the prescriber, dispenser, and patient, as well as the drug, dose, and amount dispensed. In a few states with special prescription forms, the pharmacists also capture a unique serial number that can be tracked to identify duplicates and stolen forms. PDMP information potentially allows state personnel to identify individuals who might be prescribing, dispensing, or using prescribed controlled substances inappropriately. Depending on the legally sanctioned uses of the data obtained, PDMPs can then employ various interventions designed to reduce the abuse and/or diversion of controlled substances and associated negative social and health consequences, such as drug addiction and drug overdoses [8].

Despite the increasing state and federal funding being made available to PDMPs and the program activity already underway, few empirical studies have addressed the effect of PDMPs on the prescribing or abusing of opioid analgesics [13]. Researchers have evaluated PDMPs' effect on substance abuse treatment rates from 1997 to 2003 [14] and on the prescribing of Schedule II opioid analgesics [14,15]. No known recent study has systematically evaluated the association of PDMPs with what is arguably one of the most severe consequences of opioid abuse, inadvertent lethal overdose. Accordingly, we evaluated the association of PDMPs with drug overdose mortality rates and consumption of prescription opioid medications in the United States during 1999–2005.


U.S. mortality data by state and by year for 1999–2005 were obtained from multiple cause of death mortality files produced by the National Center for Health Statistics (NCHS) at the Centers for Disease Control and Prevention (CDC). We examined drug overdose deaths that were unintentional or of undetermined intent (International Classification of Disease, 10th revision [ICD-10] codes X40–X44, Y10–Y14) and the subset of those deaths where an opioid analgesic was listed as a contributing cause of death (“opioid-related mortality”). Opioid analgesic poisoning was identified by the presence of the ICD-10 codes T40.2, T40.3, or T40.4. Overdose deaths of undetermined intent were included because some state medical examiners frequently use the undetermined intent category, and undetermined overdose deaths resemble unintentional overdoses more than suicidal overdoses [16,17].

Bridged-race census and intercensal year-specific population estimates of the 50 states and the District of Columbia (DC), developed jointly by the U.S. Census Bureau and NCHS in 2006, were obtained from the CDC Wide-ranging OnLine Data for Epidemiologic Research (WONDER) system for use in rate calculations [18]. From the Census Bureau, the authors obtained to test as covariates the median age of the population and the percentages that were Hispanic, white, black, Asian or Pacific Islander, and American Indian or Alaska Native. The Census Bureau also provided the median household income [19] and the percentages of high school and college graduates by state and year [20]. As an additional possible covariate, the authors obtained the proportions of state populations living in counties assigned by NCHS to each of four levels of urbanization [21] (of the six urbanization levels used by NCHS, the three most rural levels were combined into one to avoid small cell sizes).

State- and year-specific retail distributions of prescription opioids are tracked by the Automation of Reports and Consolidated Orders System (ARCOS) of the U.S. Drug Enforcement Administration (DEA). ARCOS monitors the flow of controlled substances from the point of manufacture through commercial distribution channels to the point of sale or distribution at the dispensing/retail level. State- and year-specific quantities of seven of the most commonly prescribed opioid drugs (fentanyl, hydrocodone, hydromorphone, meperidine, methadone, morphine, and oxycodone) were available for 1999 and 2001–2005 [6]. The DEA provided the lead author the quantities of these seven opioids for the year 2000 because they were not available on the ARCOS Website. DEA has assigned virtually all Food and Drug Administration (FDA)-approved formulations of six of these opioids to Schedule II of its Schedule of Controlled Substances, the most closely regulated of the four schedules to which prescription drugs are assigned (II–V). Hydrocodone has been sold almost entirely in combination products assigned to Schedule III, for which there are fewer restrictions on refills, documentation, and other aspects of drug dispensing [22].

To adjust for differences in opioid potency, the authors calculated morphine milligram equivalents (MME) as the product of the milligram weight of each drug and the following drug-specific multipliers: fentanyl, 75; hydrocodone, 1; hydromorphone, 4; meperidine, 0.1; methadone, 7.5; oxycodone, 1 [23]. The total MME per person and the MME per person for hydrocodone and the other six drugs were calculated separately for each state for each year.

For each of the seven study years and 51 jurisdictions (50 states and DC), a total of 357 state-years of observation, the authors determined the presence or absence of an operational PDMP. “Operational” was defined as “capable of collecting data and distributing data to one or more authorized users of the data”[24]. If a PDMP involved a major geographic subdivision but not the entire state (e.g., Virginia during 2003–2005), the authors considered it operational. If a PDMP was limited to specific prescribers who were being monitored, e.g., Washington State, it was not considered operational. Nineteen states had operational PDMPs at some time during 1999–2005: California, Hawaii, Idaho, Illinois, Indiana, Kentucky, Maine, Massachusetts, Michigan, Nevada, New Mexico, New York, Oklahoma, Pennsylvania, Texas, Utah, Virginia, West Virginia, and Wyoming.

“Proactive” PDMPs were defined as those generating reports for prescribers, dispensers, or law enforcement authorities without being solicited. PDMP reporting activity came from a 2006 survey by the Integrated Justice Information Systems Institute [25]. Thirteen of the operational PDMPs met the proactivity definition: California, Hawaii, Idaho, Maine, Massachusetts, Nevada, New Mexico, New York, Oklahoma, Pennsylvania, Texas, Utah, and Wyoming. The authors also used Integrated Justice Information Systems (IJIS) survey data to separately examine the state-years with more than 100 solicited or unsolicited reports per 10,000 population to doctors, dispensers, or other recipients. That high reporting rate was reached by four states: Kentucky, Nevada, Utah, and West Virginia [25].

The authors calculated and plotted crude mean mortality and MME rates and their standard errors for PDMP and non-PDMP states. To compare year-specific differences in means, Student's t-test was used.

The effects of PDMPs over time on 1) drug overdose mortality, 2) opioid overdose-related mortality, and 3) MME were modeled using regression models for multiple parallel time series, often referred to as “panel regression”[26]. These models included the following: structural variables (time, geography [state]), temporally lagged values of the dependent variable, and algebraically transformed (weighted), geographically lagged values of the dependent variable; the presence of a PDMP; and the potential covariates mentioned earlier added one at a time. When mortality was a dependent variable, the MME rate was tested as a potential covariate (confounder).

To prevent spatial autocorrelation (the tendency for one state to have values similar to neighboring states) from biasing estimation of the PDMP regression coefficient or its variance, we used geographically “lagged” values of the dependent variables, those related to the values of other states (weighted by proximity). Similar effects resulting from temporal autocorrelation were handled by using temporally lagged values, i.e., the values of previous years for a given rate. Because of the high level of temporal autocorrelation present in the mortality rates, these variables were also transformed by differencing. That is, instead of using the rates themselves as dependent variables, we used the year-to-year increase (or if negative, the decrease) in the rate's value. MME rates were similarly modeled as a function of the presence of a PDMP and the covariates (but not as a function of the mortality variables).

The fit of the final fixed effects panel regression model was evaluated by means of visual inspection of plots and tabulations of model residuals, and diagnostic testing of the model with Moran's I [27], Geary's C [28], and the extension of the Durbin–Watson statistic for panel data proposed by Bhargava et al. for temporal autocorrelation in multiple parallel time series [29]. In addition, techniques described by Arellano were used to mitigate the effects of heteroscedasticity (nonconstancy of the variance) in the data [30].

Substantial distortion caused by autocorrelation and indicated by regression diagnostics occurred only in the analysis of the differenced drug overdose mortality variable, the apparent result of extreme year-to-year variability in the values in drug overdose mortality rates reported by DC. Thus, indicator variables for the outlier years from DC were added to the model, resulting in satisfactory improvement in the regression diagnostics.

Where the Hausman m-statistic [26] indicated that modeling allowing for random effects was viable, we attempted such models, but no gain was noted over the fixed effects models, which permitted better diagnostic evaluation of regression modeling. Accordingly, we used fixed effects models as our final models throughout the analysis.


For states with and without PDMPs, the mean drug overdose and opioid-related overdose mortality rates rose substantially and consistently during 1999–2005. The rates approximately doubled for drug overdose mortality and tripled for opioid-related overdose mortality (Figure 1). The differences between PDMP and non-PDMP states were not statistically significant for either mortality rate for any of the study years by Student's t-test. States with PDMPs had higher crude mortality rates during 1999–2005 (Table 1).

Figure 1.

Mean drug overdose and opioid overdose mortality rates for PDMP and non-PDMP states by year, 1999–2005. Error bars indicate ±1 standard error of the mean.

Table 1.  Crude rates for drug overdose mortality, opioid overdose mortality, and morphine milligram equivalents (MME), 1999–2005
State-yearsNDrug overdose mortality rate*SEOpioid overdose mortality Rate*SEMME per personSE
  • * 

    Rate per 100,000 person years.

  • † 

    Standard errors are unadjusted for autocorrelation among observations and are therefore unsuitable for statistical tests of between-group differences.

  • ‡ 

    Morphine milligram equivalents per person per year.

  • PDMP = prescription drug monitoring program; SE = standard error.

Without PDMPs2476.460.212.200.10341.6710.20
With PDMPs1107.450.313.130.25362.4315.99
With proactive PDMPs727.640.383.300.29365.6720.47
With high-reporting PDMPs1211.410.826.570.70540.7545.54
California, New York, and Texas215.360.311.650.17251.1918.38

Proactive PDMP states did not have rates lower than other PDMP states regarding either drug overdose or opioid-related mortality nor did the states with PDMPs that sent out a high rate of reports differ from other states (Table 1). However, inspection of data for individual states revealed distinctly lower than average crude rates of drug and opioid overdose mortality on a year-by-year basis in the PDMP states of California, New York, and Texas (Table 1).

From 1999–2005, mean MME rates approximately tripled, increasing from about 175 MME/person to about 525 MME/person. PDMP and non-PDMP states had almost identical mean MME rates each year and over the entire time period (Table 1). Proactive states and states with high reporting rates did not have lower MME rates any year. As was true for mortality rates, mean MME rates in California, New York, and Texas did not increase as much as in other states. MME rates from hydrocodone were significantly higher by about 20 MME/person in PDMP states compared with non-PDMP states, whereas MME rates for the other opioids were consistently but not significantly lower by about 20 MME/person in PDMP states (Figure 2).

Figure 2.

Mean morphine milligram equivalents (MME) per person for hydrocodone and Schedule II (CII) opioids in PDMP and non-PDMP states by year, 1999–2005. Error bars indicate ±1 standard error of the mean.

In the regression analysis, the presence of a PDMP was not a significant predictor of mortality or MME rates (Table 2). However, the negative regression coefficient for PDMP in the MME model indicated a nonstatistically significant trend toward lower rates of increase in MME rates in states and years in which a PDMP existed. Adding the racial, ethnic, median age, urbanization, education, income, and (for the mortality regression analyses) MME rate variables to each of the three final regression models did not significantly decrease any of the regression coefficients for the PDMP variable so none of these variables were included in the final analysis.

Table 2.  Final models for drug overdose mortality, opioid overdose mortality, and morphine milligram equivalent rates (MME), 1999–2005
VariableDrug overdose mortality rate*Opioid overdose mortality rate*MME per person year
NameCoefficient (P)Coefficient (P)Coefficient (P)
  • * 

    Rate per 100,000 person years.

  • † 

    Morphine milligram equivalents per person per year.

  • DC00 through DC04 are indicator variables for the District of Columbia mortality rates.

Intercept0.58 (0.0001)0.23 (0.0001)3,828.99 (0.0001)
Year1–0.12 (0.6265)0.03 (0.8580)1,219.08 (0.0299)
Year20.14 (0.5784)0.16 (0.3389)1,668.44 (0.0031)
Year30.56 (0.0277)0.34 (0.0454)2,693.59 (0.0001)
Year40.30 (0.2376)0.27 (0.1125)4,560.05 (0.0001)
Year5–0.32 (0.2045)0.12 (0.4666)2,510.28 (0.0001)
DC004.42 (0.0001)  
DC02–6.88 (0.0001)  
DC035.72 (0.0001)  
DC04–3.08 (0.0001)  
Prescription drug monitoring program0.10 (0.4953)0.09 (0.3437)–162.40 (0.5535)


PDMPs were not associated with lower drug overdose mortality rates for any of the study years or with decreases (or even with lesser increases) in the rates of death resulting from drug overdoses. The findings also indicate that PDMPs were not associated with lower rates of consumption of opioids during 1999–2005, although they were associated with lower rates of use of Schedule II drugs. Even when focused on proactive PDMPs or programs with relatively high rates of reporting, there were no associations of PDMPs with trends in overdose deaths or opioid use.

This study's findings differ from those of Simeone [14], who reported that PDMPs were associated with significantly lower rates of use of opioids. However, Simeone examined only Schedule II opioids, omitting hydrocodone, whose combination products fall into Schedule III. The current study also found lower usage of Schedule II opioids, but this difference was compensated for by increased use of hydrocodone, a substitution effect also noted in an earlier study [31]. The effect of PDMPs on opioid prescribing can not be fairly evaluated without including hydrocodone because it constitutes a sizeable fraction of total opioid dosage (18% of the MME totals for the United States during 1999–2005). Hydrocodone is, in fact, the most prescribed drug in the United States [32,33]. Patients and providers might choose hydrocodone over other opioids because Schedule III drugs such as hydrocodone combination products have fewer restrictions when prescribed and lesser criminal penalties when abused. A few state PDMPs do not track Schedule III drugs.

Whether changes in the choice of drugs by physicians as a result of PDMPs adversely affect patients has been the subject of controversy [13]. For example, the addition of benzodiazepines to New York's PDMP in 1989 resulted in greater use of “less acceptable” sedating medications [34]. Some have argued that these substitution effects are transient or exaggerated [35]. Substitution of hydrocodone for other opioids is potentially problematic because of its combination with the potent hepatotoxin acetaminophen in many of its most popular formulations (e.g., Vicodin®). Narcotic–acetaminophen combination drugs now cause a large percentage of cases of liver failure resulting from acetaminophen poisoning [36]. Acetaminophen now causes more than half of the cases of acute liver failure in the United States, and the proportion is rising [36].

The observation that the three most populous states—California, New York, and Texas—have had lower rates of opioid prescribing and overdose mortality than other states with PDMPs in recent years has been made previously [3,15,37,38]. It has been suggested without explanation that these states' lower rates of opioid prescribing occurred because they had some of the oldest PDMPs in the country [15]. However, other states with long-established PDMPs, such as Massachusetts, Rhode Island, and Oklahoma, did not show similarly slower rates of increase in mortality or opioid use in this study. What might be different about California, New York, and Texas is their continued use of serialized tamper-resistant prescription forms, while other states have largely moved away from the use of special paper forms. In studies of the older, triplicate prescription forms, states consistently experienced decreased use of controlled substances following the introduction of such forms, with much of the decrease resulting from declines in inappropriate use [13].

Whether because of these special prescription forms or not, some aspect(s) of the programs in these three states might affect both overdose mortality and the rates at which opioids are prescribed. Given that these three states might be different in ways other than their continued use of special prescription forms (e.g., factors related to their large populations, their use of PDMP data, or the availability of heroin), firm conclusions can not be drawn about what is responsible for their lower mortality and opioid distribution rates. Moreover, the possible effectiveness of this specific aspect of the drug diversion control programs in these states was not among the a priori hypotheses of this study, another reason for cautious interpretation of lower mortality and drug distribution in these states.

The primary limitation of this study is its ecologic design. Ecologic studies can identify associations that are true at the state level but not at the individual level. This study attempted to rule out variables that may have confounded the analysis in this way by testing for the effect of several demographic variables on the definitive models. Adjustment for other factors that were more difficult to quantify, e.g., patterns of treatment, preventive measures such as changes in state regulations, or the availability of street drugs, was not possible. Therefore, this study can not rule out residual confounding that may have obscured a protective effect of PDMPs. For example, states with a predisposition toward drug abuse might have initially had higher drug overdose rates that made them more likely to establish a PDMP. Such a predisposition might, in fact, account for the higher crude drug overdose mortality rates seen in PDMP states in this study. As a result, we can not be certain that mortality rates in PDMP states would not have been even higher in the absence of a program. However, it can be said unequivocally that PDMP states did not do any better than non-PDMP states in controlling the rise in drug overdose mortality from 1999 to 2005.

Unfortunately, an alternative before–after design that would evaluate changes in state rates after the establishment of PDMPs to control for factors that may have led to PDMP legislation was not possible with available data. Information on opioid–analgesic-related mortality is only available after 1999. Drug distribution data are only available after 1997. Only a few states started PDMP data collection after 1999 and prior to 2005, thus allowing sufficient observation periods pre- and post-PDMP implementation. Such a study should be possible in a few years and would add to the evidence base on the effect of PDMPs on reductions in death rates due to opioid overdose.

Both the opioid distribution and mortality rates have sources of error. Errors in MME rates might have resulted when prescription drugs distributed to one state were sold to residents of neighboring states, as occurs when patients cross state lines or mail-order pharmacies distribute to multiple states. The extent to which such errors occur is not known. Overdose mortality rates might have been affected by state-to-state variation in the skill and thoroughness with which death investigations are conducted. Some overdoses might have been mistakenly attributed to natural causes, for example.

Overall drug mortality rates are a crude indicator of the prevalence of overdoses involving controlled substances monitored by PDMPs because overall rates include deaths from illicit drugs such as cocaine. However, by 2004, overdoses involving opioid analgesics easily outnumbered deaths involving heroin or cocaine in the United States. In addition, changes in overall drug mortality rates since 1999 likely reflect changes in prescription opioid-related mortality, which far exceeded changes in rates of heroin- and cocaine-related mortality in the United States during this time period [2]. Opioid-related overdose mortality rates as defined in this study are a more precise measure, but they likely suffer from variability among state medical examiners and coroners on the extent to which they specify drugs on the death certificates of overdose deaths. In addition, the data do not permit a distinction between overdoses among persons actually prescribed opioids and persons who obtained them without a prescription. PDMPs might have had more impact on persons who obtained their drugs by prescription.

Finally, this study could not evaluate the potential benefits other than prevention of overdose fatalities that might have resulted from PDMPs. For example, PDMPs are reportedly useful in facilitating criminal investigations [8], and they may have had salutary effects on drug diversion that could not be captured by our methods. Nor could this study evaluate the impact of other specific features of PDMPs that had not been captured in previous surveys, such as a high level of unsolicited reporting to law enforcement agencies.


The continuing epidemic of opioid overdose mortality documented in this study underlines the need for careful ongoing evaluation of public health and law enforcement programs designed to address drug diversion. Injury prevention programs usually should not be endorsed or abandoned based on a single evaluation so additional evaluation of PDMPs is required [39]. Clearly, however, PDMPs should work continuously to improve their effectiveness. All PDMPs should have the authority to monitor more than Schedule II drugs so that persons wanting to avoid scrutiny can not simply shift to lower-schedule drugs. The data collected by the PDMPs themselves about patterns of prescription use would probably be a useful way to monitor the effect of any such improvements. For example, PDMPs could track and report changes in standardized outcome measures, such as the percentage of patients seeing five or more doctors for controlled substances within the past 6 months or the receipt of multiple overlapping prescriptions [40]. In theory, PDMPs have the potential to address the problem of prescription drug overdoses, but to do so, their use of the information they collect will need to be enhanced.


Scott Serich, PhD, and John Eadie, DrPH, provided information from the IJIS survey. Robert Thomas, MS (Office of Statistics and Programming, National Center for Injury Prevention and Control, CDC), provided annual mortality data files. None of these individuals received compensation for their contributions.