Abstract
Purpose:
The purpose of the study is to describe and compare the number and characteristics of opioid-involved fatal cases captured in the National Poison Data System (NPDS) and in US death certificates.
Methods:
NPDS, which collects data on all calls to US poison control centers, and Drug-Involved Mortality (DIM), which combines information from literal text of US death certificates and National Vital Statistics Systems, were queried for opioid-involved fatal cases from 2010 to 2015. Characteristics of the two case series were compared.
Results:
DIM contained 154 016 opioid-involved overdose deaths, and NPDS contained 2524 fatal opioid exposures, a ratio of 61:1. The number of opioid deaths remained stable in NPDS but increased in DIM over the 6-year period. On average, deaths involving opioids with higher mean dosage strength (in morphine milligram equivalents) per unit among dispensed prescriptions were more likely to be captured in DIM relative to NPDS, as compared with those with a lower mean dosage strength per unit. The increase in fentanyl-related deaths seen in DIM since 2013 was not observed in NPDS.
Conclusions:
NPDS is a valuable drug safety surveillance resource due to its timeliness and drug specificity. However, it captures only a small fraction of opioid-involved fatal poisonings, and comparisons with data derived from death certificate literal text indicate that caution is warranted in making inferences about opioid-involved fatality trends over time or comparisons across opioids.
Keywords: analgesics, opioid, drug overdose, epidemiology, pharmacoepidemiology, product surveillance, postmarketing, vital statistics
1 |. INTRODUCTION
The ongoing national public health emergency of opioid overdose deaths necessitates timely, complete, and drug-specific information on opioid-involved fatal poisonings. Despite advances in opioid overdose surveillance,1 no perfect data source exists, due to issues such as incomplete capture of cases and case characteristics and delay in data availability. Two data sources that have been used for surveillance purposes are the National Poison Data System (NPDS) and the recently developed Drug Involved Mortality (DIM)* database.2–4
NPDS, managed by the American Association of Poison Control Centers (AAPCC), captures product-specific information on opioid exposures, demographics of exposed persons, and consequent medical outcomes based on calls to regional poison control centers (PCCs) in the United States.3 Data are available for analysis in near real time and have been used to monitor emerging trends, including opioid overdose deaths.3,5–7
The newly available DIM data resource has been used to describe opioid overdose deaths in the United States.2 DIM consists of National Vital Statistics System Mortality (NVSS-M) data linked with information extracted from the literal text of death certificates. NVSS-M data contain cause of death, demographic, and geographic information from all US death certificates. The underlying and multiple causes of death information captured in the NVSS-M data are categorized using the International Classification of Diseases, Tenth Revision (ICD-10). In most cases, ICD-10 groups drugs into broad categories, precluding identification of specific drugs involved in death. DIM enables identification of mentions of specific drugs involved in a death. Detailed methods on the identification of drug-involved deaths are described elsewhere.4
Although some research suggests that fatal opioid poisoning trends in NPDS data correlate well with NVSS-M data,6 fatal drug overdoses, particularly out-of-hospital deaths, often do not generate a call to PCCs.8–10 Therefore, it is widely recognized that NPDS captures only a fraction of deaths associated with drug poisonings.8–10 Furthermore, literature is limited on whether the proportion captured varies across time, drug product or class, suicide versus nonsuicide, and other factors. Previous studies assessing the comprehensiveness of NPDS fatality data compared with NVSS-M data have been limited by the lack of drug specificity in ICD codes.8–10 DIM allows identification of the specific opioid molecule(s) involved in death, when noted on the death certificate. In this study, we compare opioid-involved fatal poisonings in NPDS to DIM by opioid moiety and other case characteristics to better understand each data sources’ respective strengths and limitations for opioid overdose mortality surveillance.
2 |. METHODS
2.1 |. Fatal opioid poisonings—overall
NPDS and DIM were queried for all fatal opioid poisonings from January 2010 to December 2015. In NPDS,11 fatal opioid poisoning cases were defined as closed, human exposure cases that had “Death” as the medical outcome, and an AAPCC generic code indicating an opioid as an active ingredient. Direct report cases result from an exposure call that is followed to a medical outcome of death. Of note, our analyses did not include death, indirect report because of wide variability in NPDS inclusion of indirect reports across the study period (personal communication, Dr. Bruce Anderson, Executive Director of the Maryland Poison Center and Dr. Scott Schaeffer, Managing Director and Clinical Instructor at the Oklahoma Center for Poison & Drug Information, 2018). The published NPDS annual report contains both cases reported directly to a PCC from a patient, health care provider, or family member or indirectly from a medical examiner or external news source.12 The direct fatality cases receive comprehensive review from the PCC fatality review team, during which the case is verified for accuracy and assessed for relative contribution of toxic exposure to death, cause rank of substances involved, and degree of agreement between case fatality abstract in published report and NPDS database entry.13 In DIM, we defined fatal opioid poisoning cases as an ICD-10 underlying cause-of-death code in X40 to X44, X60 to X64, X85, or Y10 to Y1414 and a mention of involvement of any of the opioid terms listed in Table 1 (DIM data accessed September 20, 2018); the definition of drug involvement in DIM data has been defined elsewhere.4
TABLE 1.
Total opioid-involved fatal cases in NPDS and DIM from 2010 to 2015
| Overall Opioid-Involved Fatal Cases | |||
|---|---|---|---|
| NPDSa | DIMb | DIM/NPDS | |
| 2010 | |||
| Total cases | 456 | 20 062 | 44.0 |
| % Multisubstance | 78.5 | 56.7 | |
| 2011 | |||
| Total cases | 419 | 22 918 | 54.7 |
| % Multisubstance | 79.5 | 56.9 | |
| 2012 | |||
| Total cases | 429 | 23 525 | 54.8 |
| % Multisubstance | 76.0 | 56.0 | |
| 2013 | |||
| Total cases | 418 | 25 160 | 60.2 |
| % Multisubstance | 75.1 | 57.0 | |
| 2014 | |||
| Total cases | 384 | 28 944 | 75.4 |
| % Multisubstance | 76.3 | 60.1 | |
| 2015 | |||
| Total cases | 418 | 33 407 | 79.9 |
| % Multisubstance | 76.6 | 62.8 | |
| Total 2010–2015 | |||
| Total cases | 2524 | 154 016 | 61.0 |
| % Multisubstance | 77.0 | 58.6 | |
Abbreviations: DIM, Drug-Involved Mortality; NPDS, National Poison Data System.
NPDS cases were selected using NPDS generic codes.
DIM cases with mention of any of the following terms and an underlying cause-of-death code indicating poisoning were included: beta-hydroxy thiofentanyl, 4-fluorobutyrfentanyl, 6-acetylmorphine, acetylfentanyl, buprenorphine, butorphanol, butyrfentanyl, codeine, depropionylfentanyl, diamorphine, dihydrocodeine, ethylmorphine, fentanyl, furanylfentanyl, hydrocodone, hydromorphone, levomethorphan, levorphanol, meperidine, methadone, morphine, morphine type alkaloid, nalbuphine, norbuprenor-phine, norcodeine, norfentanyl, norhydrocodone, normeperidine, noroxyco-done, norpropoxyphene, nortramadol, O-desmethyltramadol, opiate, opioid, opioid agonist antagonist, opium, oxycodone, oxymorphone, pentazocine, propoxyphene, remifentanil, sufentanil, tapentadol, thebaine, tramadol.
We compared the total number of opioid-involved fatal cases in NPDS and DIM data from 2010 to 2015. We also examined whether these cases involved multiple substances. In NPDS, fatalities were determined to be multisubstance based on number of AAPCC generic/product code combinations or AAPCC generic codes listed in the substance variable. In DIM, multisubstance involvement was determined by the number of specific substances mentioned. Of note, NPDS captures combination products as a single substance, whereas in DIM, each component is listed as a separate substance.
2.2 |. Fatal opioid poisonings—10 most commonly dispensed opioids
We used the IQVIA National Prescription Audit™ (NPA) database to identify the ten most commonly dispensed opioids from US outpatient retail pharmacies in 2015 (Table 2). One of these opioids was fentanyl, which was separated from other opioid products because a large percentage of fatal fentanyl poisonings are believed to involve illicit fentanyl.15 We analyzed fatal poisoning cases involving these opioids, as well as heroin, in NPDS and DIM data for years 2010 to 2015. For NPDS, these 10 opioids plus heroin were identified using the active ingredient search in the Micromedex, IBM Watson Health, Greenwood Village, Colorado, USA. Micromedex codes, AAPCC generic codes, and all AAPCC product codes housed under those specific AAPCC generic codes were used to capture the active ingredient of interest, including any combination products that contained these opioids as an active ingredient.
TABLE 2.
Fatalities involving buprenorphine, codeine, hydrocodone, hydromorphone, methadone, morphine, oxycodone, oxymorphone, tramadol, heroin, or fentanyl in NPDS and DIM, 2010 to 2015
| Fatal Cases Involving Buprenorphine, Codeine, Hydrocodone, Hydromorphone, Methadone, Morphine, Oxycodone, Oxymorphone, andTramadol | Fatal Cases Involving Heroin | Fatal Cases Involving Fentanyl | ||||
|---|---|---|---|---|---|---|
| DIM | NPDS | DIM | NPDS | DIM | NPDS | |
| Total number of mortality cases: | 93 294 | 1972 | 46 259 | 296 | 18 910 | 144 |
| Age | ||||||
| Mean | 43.5 | 46.1 | 37.57 | 33.3 | 40.6 | 40.5 |
| Median | 44.0 | 47.0 | 35.0 | 31.0 | 40.0 | 37.5 |
| Age group | N (%) | N (%) | N (%) | |||
| 0–5 | 235 (0.3) | 28 (1.4) | 7 (<0.1) | 0 (0.0) | 17 (0.1) | 1 (0.7) |
| 6–12 | 47 (0.1) | 7 (0.4) | 0 (0.0) | 0 (0.0) | 4 (<0.1) | 0 (0.0) |
| 13–19 | 1653 (1.8) | 80 (4.1) | 977 (2.1) | 12 (4.1) | 281 (1.5) | 3 (2.1) |
| 20–39 | 34 210 (36.7) | 586 (29.7) | 26 464 (57.2) | 199 (67.2) | 9027 (47.7) | 68 (47.2) |
| 40–59 | 47 877 (51.3) | 831 (42.1) | 16 908 (36.6) | 66 (22.3) | 8364 (44.2) | 50 (34.7) |
| 60+ | 9264 (9.9) | 411 (20.8) | 1899 (4.1) | 7 (2.4) | 1216 (6.4) | 16 (11.1) |
| Unknown | 8 (<0.1) | 29 (1.5) | 4 (<0.1) | 12 (4.1) | 1 (<0.1) | 6 (4.2) |
| Gender | ||||||
| Male | 53 803 (57.7) | 834 (42.3) | 36 105 (78.0) | 195 (65.9) | 12 321 (65.2) | 73 (50.7) |
| Female | 39 491 (42.3) | 1136 (57.6) | 10 154 (22.0) | 101 (34.1) | 6589 (34.8) | 70 (48.6) |
| Unknown | 0 (0.0) | 2 (0.1) | 0 (0.0) | 0 (0.0) | 0 (0.0) | 1 (0.7) |
| Opioid type | ||||||
| Buprenorphine | 1453 (1.6) | 53 (2.7) | ||||
| Codeine | 3018 (3.2) | 97 (4.9) | ||||
| Hydrocodone | 18 566 (19.9) | 765 (38.8) | ||||
| Hydromorphone | 3375 (3.6) | 70 (3.5) | ||||
| Methadone | 23 614 (25.3) | 315 (16.0) | ||||
| Morphine | 21 758 (23.3) | 208 (10.5) | ||||
| Oxycodone | 32 220 (34.5) | 585 (29.7) | ||||
| Oxymorphone | 5223 (5.6) | 48 (2.4) | ||||
| Tramadol | 5815 (6.2) | 199 (10.1) | ||||
| Route | ||||||
| Oral | 31 052 (33.3) | 1807 (91.6) | 8535 (18.5) | 100 (33.8) | 3275 (17.3) | 97 (67.4) |
| Insufflation | 363 (0.4) | 47 (2.4) | 193 (0.4) | 21 (7.1) | 73 (0.4) | 15 (10.4) |
| Injection | 1956 (2.1) | 66 (3.3) | 3979 (8.6) | 150 (50.7) | 696 (3.7) | 18 (12.5) |
| Missing/other | 60 971 (65.4) | 123 (6.2) | 34 520 (74.6) | 81 (27.4) | 14 993 (79.3) | 23 (16.0) |
| Geographic region | ||||||
| Northeast | 13 934 (14.9) | 314 (15.9) | 11 535 (24.9) | 78 (26.4) | 5114 (27.0) | 29 (20.1) |
| Midwest | 16 721 (17.9) | 457 (23.2) | 14 831 (32.1) | 95 (32.1) | 4961 (26.6) | 37 (25.7) |
| South | 39 762 (42.6) | 730 (37.0) | 11 617 (25.1) | 75 (25.3) | 6901 (36.5) | 57 (39.6) |
| West | 22 877 (24.5) | 470 (23.8) | 8276 (17.9) | 47 (15.9) | 1934 (10.2) | 21 (14.6) |
| Other | 0 (0.0) | 1 (0.1) | 0 (0.0) | 1 (0.3) | 0 (0.0) | 0 (0.0) |
| Intent | ||||||
| Nonsuicide | 77 525 (83.1) | 410 (20.8) | 43 729 (94.5) | 201 (67.9) | 17 046 (90.1) | 64 (44.4) |
| Suicide | 8943 (9.6) | 1067 (54.1) | 397 (0.9) | 42 (14.2) | 579 (3.1) | 36 (25.0) |
| Unknown | 6826 (7.3) | 495 (25.1) | 2133 (4.6) | 53 (17.9) | 1285 (6.8) | 44 (30.6) |
| Year | ||||||
| 2010 | 14 419 (15.5) | 364 (18.5) | 3017 (6.5) | 24 (8.1) | 1654 (8.7) | 31 (21.5) |
| 2011 | 15 932 (17.1) | 350 (17.7) | 4563 (9.9) | 34 (11.5) | 1660 (8.8) | 19 (13.2) |
| 2012 | 15 327 (16.4) | 340 (17.2) | 6147 (13.3) | 45 (15.2) | 1613 (8.5) | 19 (13.2) |
| 2013 | 15 077 (16.2) | 316 (16.0) | 8409 (18.2) | 59 (19.9) | 1894 (10.0) | 25 (17.4) |
| 2014 | 16 040 (17.2) | 288 (14.6) | 10 850 (23.5) | 60 (20.3) | 4189 (22.2) | 18 (12.5) |
| 2015 | 16 499 (17.7) | 314 (15.9) | 13 273 (28.7) | 74 (25.0) | 7900 (41.8) | 32 (22.2) |
Abbreviations: DIM, Drug-Involved Mortality; NPDS, National Poison Data System.
We compared opioid-involved fatal cases in NPDS and DIM data for the 10 most commonly dispensed opioids and heroin for 2010 to 2015, stratified by decedents’ age, gender, opioid moiety, route of exposure, geographic region, suicide versus nonsuicide, and year. In NPDS, we defined cases with reason for exposure coded as “Intentional—suspected suicide” as suicide, “Unknown reason” and “Intentional—Unknown” as unknown, and all others as nonsuicide. In DIM, cases with underlying cause-of-death ICD-10 codes X60 to X64 were classified as suicides, cases with underlying cause-of-death ICD-10 codes Y10–14 were classified as deaths with unknown intent, and all other cases were classified as nonsuicides. Opioid route of exposure was categorized as oral, injection, insufflation, or unknown/other. A case mentioning multiple routes was counted as involving each route.
We assessed whether the ratio of opioid-involved fatal cases in NPDS to those in DIM differed by type of opioid. To explore whether state-level reporting of specific drugs on the death certificate affected the NPDS/DIM case ratio, we assessed the correlation between this ratio and statewide rate of drug-specific reporting on death certificates. We also assessed the correlation between mean opioid dosage strength per unit and ratio of DIM to NPDS cases. To estimate mean dosage strength per unit for each opioid, we used outpatient retail pharmacy dispensing data from IQVIA’s NPA database for 2015, calculating the mean morphine milligram equivalent (MME) per dosage unit (eg, tablet, capsule; MME calculated using CDC opioid oral MME conversion factors table16) among all dispensed prescriptions for that opioid. Fentanyl and buprenorphine were not included in this analysis because they also have transdermal formulations.
All data were queried independently by two analysts for each data source. SAS Enterprise Guide 7.1 was used for all analyses. In addition, an AAPCC analyst independently provided a list of generic and product codes for quality assurance. This study was reviewed and deemed to be a public health surveillance activity by an FDA IRB designee.
3 |. RESULTS
We identified 154 016 fatal opioid poisonings in DIM and 2524 fatal opioid exposures in NPDS during the study period (including illicit opioid poisonings), for an overall ratio of 61:1 (Table 1). Multisubstance cases were more common in NPDS (77%) compared with DIM (58.6%).
Restricting to the 10 most commonly prescribed opioids, there was approximately 1 NPDS direct report case to 52 cases in DIM over the 6-year time period (Figure 1). NPDS direct report cases declined slightly from 379 in 2010 to 333 in 2015, whereas cases in DIM increased from 15 574 to 22 574.
FIGURE 1.
Opioid-involved fatal cases by year for Drug-Involved Mortality (DIM), National Poison Data System (NPDS) direct report, and NPDS annual report (direct and indirect report). NPDS and DIM numbers included opioid-related fatalities with mention of buprenorphine, codeine, fentanyl, hydrocodone, hydromorphone, morphine, methadone, oxycodone, oxymorphone, or tramadol, 2010 to 2015. NPDS annual report fatality numbers include cases called in directly to the PCC from the patient or health care provider, as well as those called in indirectly from medical examiners and other external sources. NPDS annual report numbers exclude cases where clinical review determined the death as unlikely to have been related to the drug exposure
Excluding fentanyl, we identified 93 294 deaths in DIM versus 1972 deaths in NPDS for the most commonly dispensed opioids, or approximately 47 cases in DIM to every 1 case in NPDS (Table 2). For fentanyl, there were 18 910 and 144 fatal cases identified in DIM and NPDS, respectively, or approximately 131 cases in DIM to every 1 case in NPDS. For heroin, there were 46 259 and 296 fatal cases identified in DIM and NPDS, respectively, or approximately 156 cases in DIM to every 1 case in NPDS.
Mean and median ages were slightly higher in NPDS compared with DIM, with the youngest and oldest age categories having proportionally more representation in NPDS. NPDS had a higher proportion of female decedents relative to DIM. Cases involving buprenorphine, tramadol, hydrocodone, and codeine constituted a higher percentage of fatalities in NPDS while cases involving methadone, morphine, oxycodone, and oxymorphone accounted for a higher percentage of fatalities in DIM. The majority of cases in DIM did not include information on route of exposure, whereas greater than 90% of cases in NPDS specified route of exposure. The distribution by geographic region was similar in the two data sources. The proportion of cases categorized as suicide in NPDS was far higher than in DIM; for example, for the most commonly dispensed opioids excluding fentanyl, 54% of cases were classified as suicides in NPDS, compared with 10% in DIM. Annual trends were similar in the two data sources, with the exception of fentanyl, where the rapid increase in cases seen in DIM was not observed in NPDS. Characteristics of opioid fatalities in NPDS and DIM are displayed in Table 2.
The percentage of multisubstance cases (including nonopioid drugs) was generally high across the opioid moieties in both DIM and NPDS, although the proportion of multisubstance cases was higher in NPDS, overall, compared with DIM (Figure 2). The percentage of cases with suicidal intent was much higher in NPDS compared with DIM across all opioids, although the rank order of opioids, from highest to lowest percentage of cases with suicidal intent, was similar in the two data sources, with tramadol having the highest percentage of suicide cases, and fentanyl having the lowest (Figure 3).
FIGURE 2.
Percent multisubstance and single-substance fatalities involving buprenorphine, codeine, hydrocodone, hydromorphone, fentanyl, methadone, morphine, oxycodone, oxymorphone, or tramadol from 2010 to 2015 for (A) National Poison Data System (NPDS) and (B) Drug-Involved Mortality (DIM)
FIGURE 3.
Percent of fatalities involving buprenorphine, codeine, hydrocodone, hydromorphone, fentanyl, methadone, morphine, oxycodone, oxymorphone, or tramadol by intent from 2010 to 2015 for (A) National Poison Data System (NPDS) and (B) Drug-Involved Mortality (DIM)
During the study period, fentanyl, oxymorphone, morphine, and methadone had the highest DIM/NPDS case ratios, and codeine, tramadol, buprenorphine, and hydrocodone had the lowest. The DIM/NPDS case ratio for fentanyl increased sharply in 2014 to 2015, indicating that the increase in fentanyl-involved deaths observed in DIM was not observed to the same degree in NPDS. The ratio of DIM/NPDS opioid-involved fatalities by opioid molecule and year is shown in Figure 4.
FIGURE 4.
Ratio of Drug-Involved Mortality (DIM)/National Poison Data System (NPDS) opioid-involved fatalities by opioid molecule and year. Cases involving buprenorphine, codeine, hydrocodone, hydromorphone, fentanyl, methadone, morphine, oxycodone, oxymorphone, or tramadol were analyzed in DIM and NPDS, and ratio of DIM/NPDS cases is plotted by year. Inset: ratio of buprenorphine-, codeine-, hydrocodone-, hydromorphone-, fentanyl-, methadone-, morphine-, oxycodone-, oxymorphone-, and tramadol-involved cases by molecule in DIM/NPDS, 2010 to 2015
We observed a strong positive correlation between the mean dosage strength per unit among dispensed prescriptions for each opioid in 2015 and the ratio of DIM to NPDS cases involving that opioid (Figure 5). Figure S1s suggests a weak positive correlation between the percentage of drug poisoning fatalities with a specific drug mentioned on the death certificate for each state and the ratio of DIM to NPDS cases in that state. However, the variability in the DIM/NPDS case ratio increases as the rate of reporting on specific drugs on death certificates approaches 100%, ranging from less than 20 to more than 200.
FIGURE 5.
IQVIA data were used to calculate average morphine milligram equivalent (MME)/unit by prescription in 2015. These averages are plotted on the x-axis, and ratios of cases in Drug-Involved Mortality (DIM) to cases in National Poison Data System (NPDS) by opioid molecule in 2015 are plotted on the y-axis. Buprenorphine and fentanyl were excluded from this analysis because of their transdermal formulations. Methadone was included although retail dispensing data do not capture methadone dispensed directly from opioid treatment programs
4 |. DISCUSSION
We present a descriptive overview of opioid-related fatalities in NPDS and DIM to further understand differences in the total numbers, trends, and characteristics of cases captured in these two data systems. NPDS is valuable for its near real-time information and detailed, standardized information on the drugs involved in poisonings. DIM requires more time for data to become available for analysis, but captures a much higher proportion of drug-involved deaths occurring in the United States. As observed in previous research,8–10 NPDS captures only a small fraction of fatal opioid poisonings—here approximately 1 case for every 61 cases identified based on death certificate drug mentions. Furthermore, this fraction varies across drugs and over time.
In the case of opioid overdose, there are many factors that can influence whether a fatality is recorded in NPDS. As with all drug exposures, a fatal opioid exposure being captured in NPDS is contingent upon a call being generated to a PCC. A more complex case, such as a multiproduct exposure where medical advice is necessary, might be more likely to generate an exposure call than a less complex, single-product exposure. Length of survival of the exposed person could also influence the likelihood of a case being recorded in NPDS. An overdose death that occurs quickly might not be captured in NPDS if there is insufficient time to reach medical attention and generate a call to a PCC. Using the newly available DIM data source, we found that fatalities involving opioids with higher mean dosage strength per unit among dispensed prescriptions, such as morphine, oxymorphone, and methadone, are less well captured in NPDS relative to DIM compared with those involving opioids with a lower mean dosage strength per unit. We hypothesize that this difference may reflect the greater likelihood of out-of-hospital deaths with high-dosage strength opioids, with these deaths being unlikely to generate a call to a PCC. The interpretation of this finding, however, is complicated by a number of factors. First, lower dosage strength opioids like hydrocodone and codeine being primarily marketed as combination products with acetaminophen, which, when involved in an overdose, may encourage calls to poison control centers due to the known toxicity of acetaminophen. Second, the multisubstance nature of many overdoses could confound the relationship between the dosage strength and the relative likelihood of the case being captured in these two data systems.
We found that, compared with other opioids, fentanyl-related mortality was not well captured in NPDS relative to DIM, particularly in 2014 to 2015. This finding is consistent with an article published in 2018 demonstrating that mortality rates for synthetic opioids diverged for NVSS-M and poison control call data around 2012, when fentanyl-related deaths began to rise sharply.17 A separate 2017 analysis of PCC exposure calls in Chicago during a fentanyl-laced heroin overdose outbreak also did not find a spike in PCC calls in Chicago for heroin or fentanyl.18 Our comparison of NPDS to death certificate mentions of fentanyl, specifically, using DIM, further emphasizes the limitations of NPDS for real-time surveillance of fatal opioid poisonings in this evolving epidemic. Again, this may be in part due to the rapid and highly lethal respiratory depressant effects of fentanyl, which likely cause more out-of-hospital deaths and reduce the likelihood of a call to a PCC.
The smaller fraction of multisubstance opioid-involved fatalities captured in DIM compared with NPDS likely reflects the nature of NPDS data collection, which is structured to capture every drug involved in the exposure with the greatest possible granularity. In contrast, death certificates are intended to capture only the drugs that contributed to the death.19 On the other hand, cases with a single combination product may be coded as two separate substances in DIM, whereas they would be coded as a single combination substance in NPDS. This may explain the higher percentage of multisubstance cases for hydrocodone and codeine in DIM compared with NPDS, as these opioids are commonly available and dispensed in combination products with acetaminophen. Cases involving heroin and fentanyl had a lower percentage of multisubstance cases, possibly reflecting their high potential for fatal poisoning regardless of concomitant substances, as well as perhaps a lower likelihood of additional substances being implicated by the certifier as contributing to the death.
While DIM captured more fatal opioid poisonings overall, NPDS contained more information on route. Whereas the majority of NPDS cases were classified as oral route, the majority of DIM cases were classified as unknown/other route. It is important to note, however, that route of exposure is not linked to a specific product in multisubstance NPDS cases but instead is assigned at the level of the case. For example, a case involving multiple opioid products administered through different routes may have documented more than one route, but the routes would not be linked to the specific opioid products. NPDS also contains detailed information at the level of the product(s) involved in the exposure, while DIM has information at the level of the opioid moiety only. Linking these two data sources could provide more comprehensive information on total number of deaths caused by each opioid moiety and which specific products were involved.
In this analysis, only 1 in 10 fatal opioid poisonings in DIM was classified as a suicide, whereas more than half in NPDS were. Previous work has noted that when clinicians independently reviewed opioid-involved intentional exposure cases from PCCs, they were less likely to categorize the case as suicidal intent than the PCC specialist initially coding the case.20 In addition, the National Association of Medical Examiners suggests requiring a “preponderance of evidence” when classifying a death as a suicide,21 and medical examiners may be reluctant to rule a death as suicide in order to avoid stigma or pain for the surviving family.22 However, the rank of opioids by proportion of cases classified as suicide was relatively consistent between DIM and NPDS. Another possibility is that this is reflective of the types of calls that are reported to NPDS and that suicide cases are more likely to generate a call to a PCC.
Previous studies have compared data from NPDS to vital statistics,8,9,23,24 medical examiner,9,18,25 hospital,10,23 emergency department visit,7,26 and emergency medical service record data.18 Some found a strong correlation between NPDS data and the secondary data source, whereas others showed poor correlation. This variation could be due to several factors, including geographic differences, time period, and the outcomes and drugs examined. For example, we found that even in states with a high level of drug-specific reporting, the ratio of DIM/NPDS cases was extremely variable, indicating high geographic variability in NPDS capture of fatal opioid poisonings. We also found that both the drug and time period influenced the relative number of fatal poisonings documented in PCC versus death certificate text data.
Our study has several limitations. First, DIM data are derived from the literal text written on death certificates. While more comprehensive than NPDS data, DIM will not consistently classify every opioid-involved fatal poisoning, because documentation of specific drug involvement, suicide versus nonsuicide, and other case elements depends heavily on jurisdictional practices in death investigation and documentation. These practices may change over time.27,28 This could be particularly important in the case of fentanyl, where diagnostic practices may have changed over time to specifically test for the presence of the drug more often as fentanyl-related fatalities have risen. Therefore, DIM cannot be considered a gold standard, and the results of this study should be interpreted with these limitations in mind. Second, it is possible that morphine and heroin cases are misclassified in one or both data sources. Heroin (diamorphine) is metabolized into 6-monoacetylmorphine and subsequently to morphine.29 Therefore, toxicology tests may misidentify morphine in cases of heroin exposure. Third, our multisubstance quantification is likely an underestimate for some molecules in DIM because this variable does not include mentions of involvement of drug classes (eg, benzodiazepine) without mention of a specific substance (eg, alprazolam) and an overestimate in others that are components of a combination product, as mentioned previously. Fourth, as new products, licit or illicit, become available, naming conventions and categorization of these products changes in NPDS and DIM. We endeavored to create the most comprehensive list of products and metabolites in each of these databases, but it is possible that less common moieties were inadvertently excluded from our overall opioid-involved mortality numbers. Fifth, products that were not in Micromedex or housed under a generic code for our opioids of interest were not captured in our NPDS query. However, the numbers of cases for our opioids of interest in the 2010–2015 AAPCC annual reports suggest that our analysis missed fewer than 1% of all cases, mostly hydrocodone and hydromorphone cases coded as “unknown or other narcotic.”
In conclusion, NPDS and DIM are both valuable data sources for studying and monitoring opioid-involved fatal poisonings; however, it is important to understand their differences and limitations, particularly with regard to estimating national trends for fatal poisonings involving specific opioids. The DIM data source, while it has more complete case capture, lacks much of the detailed information found in NPDS, while NPDS’ well-recognized underascertainment of fatal poisonings may differ systematically by opioid, year, and other case characteristics. Further research linking NPDS cases to DIM cases by decedent would improve our understanding of the representation of specific case demographics in both NPDS and DIM.
Supplementary Material
KEY POINTS.
The National Poison Data System (NPDS) is a valuable drug safety surveillance resource due to its timeliness and drug specificity, but caution should be used in making inferences regarding opioid-related fatality trends over time and comparisons across opioids.
The ratio of fentanyl-related fatal poisonings captured in the Drug Involved Mortality (DIM) data source compared with NPDS is notably higher than the ratio for other opioids examined in this paper, and the discrepancy has increased over time.
The ratio of opioid-involved fatal poisonings captured in NPDS compared with those captured in the more comprehensive DIM data, which is based on death certificate literal text, varies considerably by the opioid involved and year.
A positive correlation exists between the relative likelihood of identifying fatal poisonings in DIM compared with NPDS and a higher mean dosage strength per unit (morphine milligram equivalents/unit) among dispensed prescriptions for the opioid involved.
ACKNOWLEDGEMENTS
The research presented in this paper is that of the authors and does not reflect the official policy or opinion of the US Food and Drug Administration. Support for this study came from the US Food and Drug Administration. This project was supported, in part, by an appointment to the Research Participating Program at the Center for Drug Evaluation Research, administered by the Oak Ridge Institute for Science and Education (ORISE) through an interagency agreement between the US Department of Energy and the US Food and Drug Administration. ORISE had no role in the study design, review, or decision to publish.
We would like to thank Jennie Wong from the Office of Surveillance and Epidemiology, Food and Drug Administration for her kind support in part of the analyses. We would also like to thank Dr Scott Schaeffer, managing director of the Oklahoma Center for Poison & Drug |Information, and Dr Bruce Anderson, Executive Director of the Maryland Poison Center, for providing information on National Poison Data System indirect report cases.
Funding information
U.S. Food and Drug Administration; Oak Ridge Institute for Science and Education; US Food and Drug Administration
Footnotes
The DIM data were created using Drug Mentioned in Involvement (DMI) methodology to identify each mention of a drug which was categorized as being involved in death.
ETHICS STATEMENT
This study was reviewed and deemed to be public health surveillance activity by an FDA IRB designee
CONFLICT OF INTEREST
The authors declare no conflict of interest.
SUPPORTING INFORMATION
Additional supporting information may be found online in the Supporting Information section at the end of the article.
REFERENCES
- 1.CDC. Enhanced State Opioid Overdose Surveillance. 2017. 12/16/2018]; Available from: https://www.cdc.gov/drugoverdose/foa/state-opioid-mm.html.
- 2.Warner M, Trinidad JP, Bastian BA, Miniño AM, Hedegaard H. Drugs most frequently involved in drug overdose deaths: United States, 2010–2014. Natl Vital Stat Rep. 2016;65(10):1–15. [PubMed] [Google Scholar]
- 3.Mowry JB, Spyker DA, Brooks DE, Zimmerman A, Schauben JL. 2015 Annual Report of the American Association of Poison Control Centers’ National Poison Data System (NPDS): 33rd Annual Report. Clin Toxicol (Phila). 2016;54(10):924–1109. [DOI] [PubMed] [Google Scholar]
- 4.Trinidad JP, Warner M, Bastian BA, Miniño AM, Hedegaard H. Using literal text from the death certificate to enhance mortality statistics: characterizing drug involvement in deaths. Natl Vital Stat Rep. 2016;65(9):1–15. [PubMed] [Google Scholar]
- 5.RADARS® System Response to Docket No. FDA-2017-N-2903: Data and Methods for Evaluating the Impact of Opioid Formulations with Properties Designed to Deter Abuse in the Postmarketing Setting.
- 6.Ronk C, B.B., Severtson G, Green J, Dart R, Opioid-Related Mortality: National Vital Statistics System versus RADARS(R) Poison Center Program. RADARS(R) System Technical Report, 2014Q3. [Google Scholar]
- 7.Davis JM, Severtson SG, Bucher-Bartelson B, Dart RC. Using poison center exposure calls to predict prescription opioid abuse and misuse-related emergency department visits. Pharmacoepidemiol Drug Saf. 2014;23(1):18–25. [DOI] [PubMed] [Google Scholar]
- 8.Hoppe-Roberts JM, Lloyd LM, Chyka PA. Poisoning mortality in the United States: comparison of National Mortality Statistics and Poison Control Center Reports. Ann Emerg Med. 1999;35(5):440–448. [PubMed] [Google Scholar]
- 9.Soslow AR, Woolf AD. Reliability of data sources for poisoning deaths in Massachusetts. Am J Emerg Med. 1992;10(2):124–127. [DOI] [PubMed] [Google Scholar]
- 10.Blanc PD, Jones MR, Olson KR, Surveillance of poisoning and drug over dose through hospital discharge coding, poison contrl center reporting, and the drug abuse warning network. Am J Emerg Med, 1993. 11(1): p.14–9, 1. [DOI] [PubMed] [Google Scholar]
- 11.Gummin DD, Mowry JB, Spyker DA, Brooks DE, Fraser MO, Banner W. 2016 Annual Report of the American Association of Poison Control Centers’ National Poison Data System (NPDS): 34th Annual Report. Clin Toxicol (Phila). 2017;55(10):1072–1252. [DOI] [PubMed] [Google Scholar]
- 12.AAPCC. National Poison Data System Annual Reports. 2017. 9/10/2018]; Available from: https://www.aapcc.org/annual-reports.
- 13.Bronstein AC, Spyker DA, Cantilena LR Jr, Green JL, Rumack BH, Heard SE. 2008 Annual Report of the American Association of Poison Control Centers’ National Poison Data System (NPDS): 26th Annual Report. Clin Toxicol (Phila). 2009;47(10):911–1084. [DOI] [PubMed] [Google Scholar]
- 14.Rudd RA, Seth P, David F, Scholl L. Increases in drug and opioid-involved overdose deaths—United States, 2010–2015. MMWR Morb Mortal Wkly Rep. 2016;65(50–51):1445–1452. [DOI] [PubMed] [Google Scholar]
- 15.O’Donnell JK, Gladden RM, Seth P. Trends in deaths involving heroin and synthetic opioids excluding methadone, and law enforcement drug product reports, by census region—United States, 2006–2015. MMWR Morb Mortal Wkly Rep. 2017;66(34):897–903. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.CDC. Opioid Oral Morphine Milligram Equivalent (MME) Conversion Factors. 5/15/2018]; Available from: https://www.cms.gov/Medicare/Prescription-Drug-Coverage/PrescriptionDrugCovContra/Downloads/Opioid-Morphine-EQ-Conversion-Factors-April-2017.pdf.
- 17.Iwanicki JL, Severtson SG, Margolin Z, Dasgupta N, Green JL, Dart RC. Consistency between opioid-related mortality trends derived from Poison Center and National Vital Statistics System, United States, 2006–2016. Am J Public Health. 2018;108(12):1639–1645. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Moore PQ, Weber J, Cina S, Aks S. Syndrome surveillance of fentanyl-laced heroin outbreaks: utilization of EMS, Medical Examiner and Poison Center databases. Am J Emerg Med. 2017;35(11):1706–1708. [DOI] [PubMed] [Google Scholar]
- 19.Goldberger BA, Maxwell JC, Campbell A, Wilford BB. Uniform standards and case definitions for classifying opioid-related deaths:recommendations by a SAMHSA consensus panel. J Addict Dis.2013;32(3):231–243. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Smith MY, Dart R, Hughes A, et al. Clinician validation of poison control center (PCC) intentional exposure cases involving prescription opioids. American Journal of Drug & Alcohol Abuse. 2006;32(3):465–478. [DOI] [PubMed] [Google Scholar]
- 21.National Association of Medical Examiners. In: Hunsaker JC, Hanzlick R, Davis GH, eds. A Guide for Manner of Death Classification. Editor, Atlanta, Georgia: National Association of Medical Examiners; 2002. [Google Scholar]
- 22.Stone DM, Holland KM, Bartholow B, et al. Deciphering suicide and other manners of death associated with drug intoxication: a Centers for Disease Control and Prevention Consultation Meeting summary.Am J Public Health. 2017;107(8):1233–1239. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Bunn TL, Yu L, Spiller HA, Singleton M. Surveillance of methadone-related poisonings in Kentucky using multiple data sources.Pharmacoepidemiol Drug Saf. 2010;19(2):124–131. [DOI] [PubMed] [Google Scholar]
- 24.Dasgupta N, Davis J, Funk MJ, Dart R. Using poison center exposure calls to predict methadone poisoning deaths. PLoS ONE. 2012;7(7):1–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Friedman LS. Real-time surveillance of illicit drug overdoses using poison center data. Clin Toxicol (Phila). 2009;47(6):573–579. [DOI] [PubMed] [Google Scholar]
- 26.Lee S, Klein-Schwartz W, Welsh C, Doyon S. Medical outcomes associated with nonmedical use of methadone and buprenorphine. J Emerg Med. 2013;45(2):199–205. [DOI] [PubMed] [Google Scholar]
- 27.Ruhm CJ. Geographic variation in opioid and heroin involved drug poisoning mortality rates. Am J Prev Med. 2017;53(6):745–753. [DOI] [PubMed] [Google Scholar]
- 28.Manini AF, Nelson LS, Olsen D, Vlahov D, Hoffman RS. Medical exam iner and medical toxicologist agreement on cause of death. Forensic science international. 2011;206(1–3):71–76. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Smith ML, Shimomura ET, Summers J, et al. Urinary excretion profiles for total morphine, free morphine, and 6-acetylmorphine following smoked and intravenous heroin. J Anal Toxicol. 2001;25(7):504–514. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.