Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2024 Jan 1.
Published in final edited form as: Drug Alcohol Depend. 2022 Nov 22;242:109710. doi: 10.1016/j.drugalcdep.2022.109710

The prevalence of opioid use disorder in Kentucky’s counties: A two-year multi-sample capture-recapture analysis

Katherine Thompson 1,*, Joshua A Barocas 2,*, Chris Delcher 3,4, Jungjun Bae 3,4, Lindsey Hammerslag 5, Jianing Wang 6, Redonna Chandler 7, Jennifer Villani 7, Sharon Walsh 8,9, Jeffery Talbert 5,10
PMCID: PMC9772240  NIHMSID: NIHMS1855191  PMID: 36469995

Abstract

Background:

Kentucky has one of the highest opioid overdose mortality rates in the United States. Accurate estimates of people with opioid use disorder (OUD) are critical to plan for the scope of interventions required to reduce overdose and opioid misuse. Commonly used household surveys are known to underestimate OUD at the state-level and do not provide county-level estimates.

Methods:

We performed a multi-sample capture-recapture analysis to estimate OUD prevalence in Kentucky in 2018 and 2019. We utilized four statewide datasets that were linked at the individual level: 1) Registry of Vital Statistics, 2) Emergency Medical Services (EMS), 3) Kentucky’s Prescription Drug Monitoring Program (PDMP), and 4) Kentucky Medicaid. We included persons aged 18 to 64 years who resided in Kentucky between 2018 and 2019. We identified individuals with administrative data consistent with OUD in each of the datasets, including a fatal opioid-involved overdose (Vital Statistics), EMS runs for suspected opioid overdose, receipt of buprenorphine for OUD treatment (PDMP), or Medicaid claims for OUD. Observed and estimated counts of OUD cases and prevalence of OUD among the adult population in Kentucky.

Results:

The estimated statewide OUD prevalence was 5.5% and 5.9% for 2018 and 2019, respectively, ranging from 1.3% to 17.7% across Kentucky counties. As expected, counties with the highest OUD rates were Appalachian counties (eastern area) of the state.

Conclusions:

Our analysis reveals a substantially larger proportion of KY residents have OUD than previously estimated. Our approach offers a model for states needing county-level estimates of OUD.

Keywords: opioid use disorder, epidemiology, capture-recapture

1. Introduction

Overdose deaths due to opioids have been steadily rising in the United States for more than twenty years, with the greatest increases in the past decade. Some of the most recent available data demonstrate the U.S. experienced over 93,000 overdose deaths in 2020, the largest number of overdose deaths recorded in a single year and an increase of more than 30% from 2019 (Centers for Disease Control and Prevention, 2021). Much of the sharp rise is attributable to the use of illicitly manufactured fentanyl and other synthetic opioids among people already suffering from opioid use disorder (OUD) and the disproportionate impact of the COVID-19 pandemic on this population. Approximately 9% of 2020 overdose deaths occurred in the Ohio Valley, which includes West Virginia, Ohio, and Kentucky, a region already experiencing high rates of overdose. In 2019, Kentucky was among the top ten most highly affected states by the opioid crisis, with a drug overdose age-adjusted mortality rate of 32.5/100,000, approximately 50% higher than the national rate (21.6/100,000) (Kaiser Family Foundation, 2022). Additionally, Kentucky increase in drug-overdose deaths during the start of the COVID-19 pandemic (Slavova et al., 2021) was the second highest percentage increase in the nation (Centers for Disease Control and Prevention, 2022). Accurate and timely estimates of OUD are critical to gauge the number of people at-risk for overdose and to plan harm reduction and treatment interventions adequately matched to the scope of this problem.

Survey data, from people who self-report opioid use, and administrative data, capturing individuals with a formal OUD diagnosis, are routinely used to estimate OUD prevalence (Substance Abuse And Mental Health Services Administration, 2015; Substance Abuse and Mental Health Services Administration, 2020a). These approaches likely underestimate substantial portions of the population with OUD due to a combination of factors (Mastro et al., 1994; Origer, 2012). First, the stigma associated with substance use can lead to misclassification in surveys using self-report, due to social desirability bias (Chen et al., 2006). Second, structural barriers, such as lack of access to medications for opioid use disorder (MOUD) and harm reduction services, criminalization of drug use, and unaffordable housing prevent people who use drugs from being counted (Mastro et al., 1994; Origer, 2012). A recent study by Krawczyk et al (2022) noted that in 2018, there were 25,279 individuals who were dispensed buprenorphine and 5,004 who were dispensed MOUD at an opioid treatment program (largely methadone). In 2019, those numbers were 26,265 and 4,782, respectively. Additionally, there are 82 operating SSPs in Kentucky in 62 counties and Kentucky has a standing order naloxone prescription. Third, they often do not include county-level information, which is needed for geographically targeted interventions and service planning. Resource allocation at the county-level would be better served by improved estimates of local OUD prevalence.

Alternative approaches have been employed, particularly when additional data types are available, to obtain more robust prevalence estimates of people who use drugs (Hickman and Taylor, 2005). The estimated prevalence of OUD or other substance misuse using these alternative approaches is consistently higher than routine survey or administrative data estimates (Barocas et al., 2018; Comiskey and Barry, 2001; Khazaei et al., 2012; Wood et al., 2000). In settings where linked datasets exist, the capture-recapture/multiple systems estimation approach is a method to improve the accuracy of estimates of the prevalence of drug use in large populations and in subgroups (Bird and King, 2018). Capture-recapture methods utilize basic laws of probability to estimate the size of a population based on the overlap between groups of individuals observed in different data sets (Seber, G. A. F., 2002). These methods have been employed to assess the completeness of county-level HIV surveillance (Wesson et al., 2018), hepatitis C prevalence (Nielsen et al., 2020), and, more recently, in estimating the prevalence of OUD (Barocas et al., 2018; Hay and Richardson, 2016; Hickman et al., 1999; Hickman and Taylor, 2005).

The most recent (2019) estimates from the National Survey on Drug Use and Health (NSDUH) suggest that approximately 1.3% of the population residing in Kentucky ages 12 years and older self-reported OUD in the past year (Substance Abuse and Mental Health Services Administration, 2020b). However, we hypothesized that the true rate is higher in this socially vulnerable state, and that there is considerable variability in OUD prevalence across Kentucky’s 120 counties. Thus, our objective was to adapt capture-recapture methods previously used in Massachusetts (Barocas et al., 2018) to estimate 2018 and 2019 (the most recent available) state- and county-level prevalence of OUD in Kentucky.

2. Methods

2.1. Data Sources

Capture-recapture methods require more than one dataset containing records from the same individual. In Kentucky, the datasets available for this analysis included: Vital Statistics, Emergency Medical Services (EMS), the prescription drug monitoring program (PDMP, known as KASPER- Kentucky All-Schedule Prescription Electronic Reporting), and Medicaid. The Medicaid data included only Medicaid beneficiaries aged 18–64; the remaining three data sources included any Kentucky resident aged 18–64. Using these four datasets, a master contingency table was created by the Kentucky Cabinet for Health and Family Services Department of Medicaid Services Data Analytics Team after linking individuals across each dataset. Data cleaning and data linkage processes are described below. The data linkage process required six linked tables: Medicaid to Vital Statistics, Medicaid to EMS, Medicaid to PDMP, PDMP to EMS, PDMP to Vital Statistics, and EMS to Vital Statistics.

2.2. Data Linkage

The Kentucky Cabinet for Health and Family Services (CHFS) uses a master data management system, IBM InfoSphere® to manage Medicaid eligibility data and Vital Statistics death data (International Business Machines Corporation, 2021), and links the data using the InfoSphere® master recipient ID. Linkage of Medicaid data to Vital Statistics data is already performed by CHFS via IBM InfoSphere®. Linkage from Medicaid to EMS, Medicaid to PDMP, PDMP to Vital Statistics, EMS to Vital Statistics, and EMS to PDMP used a manual linking process.

The EMS data were also processed to create a MasterIncidentID for individuals with more than one EMS call record using match logic from social security number (SSN), date of birth (DOB), and name. Names were matched either by exact matching or by “SOUNDEX” SQL matching (National Archives, 2016). SOUNDEX converts an alphanumeric string to a four-character code that is based on how the string sounds when spoken in English, which permits matching of homophones with different spelling. The first level of matching required both SSN and DOB to match and a SOUNDEX match on first or last name. The second level of matching required SSN to match, plus a SOUNDEX match for both first and last name. The third level of matching required DOB to match, plus a SOUNDEX match for both first and last name, plus an address match.

We assigned IDs and demographic information using the hierarchal method: Vital Statistics, EMS, PDMP, and Medicaid. In particular, the hierarchical method worked as follows. If a person was identified in a single data source, the ID and demographic information from the data source were retained. Otherwise, if a person was linked across the two or more data sources, the ID and demographic information from the data sources were used in the following priority order, (omitting the data sources that the person was not observed in): Medicaid first, PDMP second, EMS third, and Vital Statistics fourth. The number of individuals observed as overlapping in each combination of data sources was counted for use in multiple systems estimation.

2.3. Inclusion/Exclusion criteria

We included persons ages 18 to 64 years within the dataset who were identified as persons with OUD within any of the linked data sources (Table 1). The definitions in this analysis were similar to those used in previous work (Barocas et al., 2018). Medicaid and Vital Statistics data contained known sex, age, and county data for each person. PDMP data excluded persons without data for our demographic categories (age, sex, or county of residence), totaling at most 1448 persons aged 18–64 in each of 2018 and 2019. EMS data excluded persons without data for our demographic categories (age, sex, or county of residence), totaling 531 and 598 persons aged 18–64 in each of 2018 and 2019, respectively. Persons identified as having OUD within any of the linked data sources comprise the “known” (or equivalenty, “observed”) counts throughout the manuscript.

Table 1.

Individual Datasets Utilized and Opioid Use Disorder Variable Definitions

Data set Description of data set OUD Variable Definition*
Medicaid Medicaid Claims Data ICD10 codes (F11.20, F11.21, F11.10, F11.120—F11.122, F11.129, F11.14, F11.150, F11.151, F11.159, F11.181, F11.182, F11.188, F11.190-F11.222, F11.229, F11.23, F11.24, F11.250, F11.251, F11.259, F11.281, F11.282, F11.288, F11.29, F11.90-F11.922, F11.929, F11.913, F11.914, F11.950, F11.951, F11.959, F11.981, F11.982, F11.989, F11.99) for opioid use disorder, OR. a buprenorphine product indicated for treatment of OUD (Supplemental Table S2).
Vital Statistics Death Certificates ICD10 codes for mortality from the underlying cause of death field to identify poisonings/overdoses: X40-X44, X60-X64, X85, and Y10-Y14. All multiple cause of death fields were then used to identify an opioid-related overdose death: T40.0, T40.1, T40.2, T40.3, T40.4, and T40.6.
PDMP Controlled Substance Drug Records A prescription fill of a buprenorphine product indicated for treatment of OUD (Supplemental Table S2).
EMS Emergency Medical Services reports Suspected opioid overdose in EMS run reports*
Open in a new tab

Abbreviations: PDMP = Prescription Drug Monitoring Program; EMS = Emergency Medical Services.

*

See Slavova et al. 2021 and Lasher et al. 2019 for full definition.

2.4. Data Analysis

  • Step 1: People with indicators of OUD (“known”) in each dataset were identified (Table 1). For example, a prescription fill of a buprenorphine product indicated was used to identify people with OUD in the PDMP dataset though methadone and naltrexone are also used to treat OUD (among other conditions such as alcohol use disorder and pain, respectively). Notably, Kentucky’s PDMP in 2018–2019 did not include any records for methadone or naltrexone prescriptions and so this dataset represented only persons receiving buprenorphine MOUD. For each year, data were either aggregated to the county-level alone or stratified by sex (male/female), age group (ages 18–34 years, 35–54 years, 55–64 years) within counties. Note that each year was analyzed separately, meaning that no matching occurred between data from 2018 and data from 2019. The master contingency table was a tabular matrix containing the counts of individuals with known OUD in each data source by stratum.

    Previous methods showed that when heterogeneity across strata is present, then strata should be analyzed separately (Bird and King, 2018; Tilling and Sterne, 1999). For the three counties with the largest population sizes (Fayette, Jefferson, and Kenton), heterogeneity was clear across strata, so we analyzed these data on the stratum level. The remaining 117 counties were analyzed on the county level because there was not heterogeneity in their OUD case counts, as shown by the lack of variability in OUD case counts across strata within counties. After analysis, we performed a sensitivity analysis to check that lack of heterogeneity was an appropriate assumption.

  • Step 2: To estimate the number of people previously unknown to have OUD (“unknown”, or people who are unobserved in the linked data sources), we employed a Multiple Systems Estimation/Capture-Recapture method designed for data sets with nonoverlap of individuals across data sources (Chan et al., 2021). This method utilizes a stepwise selection process (threshold set to 0.05) with Poisson Generalized Linear Models (GLMs), considering main effects and two-way interactions (Chan et al., 2021). We used the R package, SparseMSE to conduct these analyses (Chan et al., 2019). The sum of “known” and “unknown” OUD counts gives the estimated total OUD counts for each county.

  • Step 3: We estimated confidence intervals for county-level estimates. For county-level analyses, we obtained confidence intervals for unknown OUD case counts using the bootstrap method from prior research (Barocas et al., 2018). In our case, we deployed this method for each single county or using six strata for Fayette, Jefferson, and Kenton counties. We deployed the same method for the statewide estimates, using all 135 strata/county OUD case counts estimates in our calculations. To obtain confidence intervals for the total OUD case counts, we added the known OUD case counts to the confidence interval bounds for unknown OUD case counts.

  • Step 4: We calculated OUD prevalence (and confidence interval limits) for 2018 and 2019 by dividing the estimated total number of OUD cases (and its confidence interval limits) by the adult population (18–64 years) for the given year and county.

  • Step 5: We performed a sensitivity analysis that excluded the vital statistics dataset because deaths involving opioids may not always indicate an underlying OUD. Indeed, the stimulant supply (e.g., cocaine, methamphetamine) frequently contains fentanyl and other opioids, leading to high coinvolvement of multiple types of drugs in an overdose death (Mattson et al., 2021).

County population for individuals ages 18 to 64 years was obtained from the Centers for Disease Control and Prevention WONDER database (National Center for Health Statistics, 2021). Appalachian county status was obtained from the Appalachian Regional Commission (Appalachian Regional Commission, 2022). ArcGIS Pro 2.8.3 (Esri Inc., Redlands, California, USA) was used to create maps (Esri, 2021).

3. Results

In 2019, the number of people 18–64 years old with known OUD in Kentucky was 98,015, an increase of 6.6% from 2018 (91,938). Large counties with the greatest number of people with known OUD were Jefferson [Louisville Metro Area: 10,537 (2018); 11,382 (2019)] and Fayette [Lexington Metro Area: 5,264 (2018); 5,492 (2019)]. Perry (13.6%) and Clay (16.1%) counties had the highest known prevalence in 2018 and 2019, respectively. All county-level data are available in the Supplemental Appendix (Supplemental Table 1).

Among people with known OUD, 52% were male, more than half were 35–54 years old, one third were 18–34 years old, and one tenth were 55–64 years old, which was consistent for both years. Figure 1 shows graphically the intersection of people with known OUD in each data set. For example, in 2019, there were 78,040 people with OUD identified from records in Medicaid where 40,956 (52%) people were found in at least one of the other datasets and 37,084 (48%) just in Medicaid alone.

Figure 1.

Figure 1.

Figure 1.

Open in a new tab

Venn diagram of known cases of persons ages 18 to 64 years old in Kentucky, 2018 (top, N=91,938) and 2019 (bottom, N=98,015) by administrative dataset.

Each dataset used in the capture-recapture analysis is represented by a large oval. The total number of persons with opioid use disorder (OUD) and percentage of the total is noted under the dataset name (does not add to 100% due to overlapping records). Totals shown in the diagram sum to 100% after accounting for overlaps. For smaller counts found in the center of the diagrams, we summed and place the total in the call-out.

The estimated number of people 18–64 years old with unknown OUD in Kentucky was 56,450 and 61,326 people in 2018 and 2019, respectively. Counties with the greatest number of people previously unknown to have OUD were Jefferson (22,813, 95% CI: 22,526–23,113), Kenton (12,317, 95% CI: 12,099–12,536), and Fayette (3,920, 95% CI: 3,796–4,040) in 2018; and Jefferson (46,284, 95% CI: 45,869–46,707), Hardin (1,242, 95% CI: 1,173–1,310), and Kenton (1,135, 95% CI: 1,068–1,200) in 2019.

Summing the known and unknown counts at the state-level, the estimated total number of people with OUD was 148,388 (95% CI: 147,931–148,854) in 2018 and 159,341 (95% CI: 158,856–159,833) in 2019 – a 7.4% increase from 2018 to 2019. (Table 2). Thus, the estimated total OUD prevalence in Kentucky among people 18 to 64 years old was 5.45% (95% CI: 5.43%−5.46%) and 5.87% (95% CI: 5.85%−5.89%) in 2018 and 2019, respectively. At the county-level, the highest estimated total OUD prevalence occurred in Kenton (15.1%, 95% CI: 14.9%−15.3%) and Clay (17.7%, 95% CI: 17.5%−17.9%) counties, while Union (1.4%, 95% CI: 1.3%−1.6%) and Hancock (1.3%, 95% CI: 1.1%−1.4%) counties had the lowest prevalence for 2018 and 2019, respectively. The two most populated counties in 2019 – Jefferson (population 470,441) and Fayette (population 211,168) – had OUD prevalence estimates of 7.0% and 4.3% in 2018 and 12.2% and 2.9% in 2019, respectively. Appalachian counties had higher total OUD prevalence estimates compared to non-Appalachian counties (7.1% versus 4.9% in 2018 and 7.5% versus 5.3% in 2019, respectively). From 2018 to 2019, 31 counties experienced significant (non-overlapping CIs) increases and 29 counties significant decreases in OUD in Kentucky (Figure 2). For additional details, see Supplemental Table 1.

Table 2.

Results of Capture-Recapture Estimate of Opioid Use Disorder Prevalence in Kentucky, 2018 and 2019

Known (n) Unknown (n) Unknown 95% CI Estimate (n) Estimate 95% CI Estimated Prevalence (%) Estimated Prevalence (%) 95% CI
2018 Total 01 56,450 (55,993, 56,916) 148,388 (147,931, 148,854) 5.45 (5.43, 5.46)
2019 Total 98,015 61,326 (60,841, 61,818) 159,341 (158,856, 159,833) 5.87 (5.85, 5.89)
Open in a new tab

Abbreviations: n = number; CI = confidence interval

Figure 2.

Figure 2.

Open in a new tab

Estimated OUD prevalence (%) among 18 to 64 year olds in Kentucky counties, 2019 (base map).

The map demonstrates the estimated OUD prevalence by county in 2019. The dark black outline denotes the Appalachian region. Counties with a significant increase (n=31) in OUD between 2018 and 2019 are shown with an red arrow; counties with a significant decrease (n=29) are shown with a green arrow; other counties did not have a significant change. Cutpoints for rate categories were determined using Jenks Natural Breaks Optimization in ArcGIS.

The sensitivity analysis excluding the death records resulted in a lower overall OUD prevalence of 4.7% (19,391 fewer estimated unknown OUD cases) in 2018 and a higher prevalence of 6.7% (23,307 more estimated unknown OUD cases) in 2019. The number of known OUD cases remained similar at 91,225 in 2018 and 97,326 in 2019.

4. Discussion

This work aimed to improve OUD prevalence estimates in Kentucky for 2018 and 2019 by estimating all people with OUD that do not appear as “known” in one of the four linked data sources, in addition to providing the number of people with OUD as identified as one of the four data sources. In 2019, Kentucky had almost 160,000 people aged 18 to 64 years old with OUD—nearly 6% of the population – and county-level prevalence ranged from 1% to 18%. These findings are the first estimates of OUD prevalence available for all of Kentucky’s 120 counties. Strikingly, our full state estimate is similar to the 5% OUD rates reported from 11 other state Medicaid programs in 2018 (The Medicaid Outcomes Distributed Research Network (MODRN) et al., 2021). OUD rates surpassed 7% in Appalachian counties. Among the 23 counties in the state with increases in OUD between 2018 and 2019, 17 were located in the Appalachian region. Our findings of the strong geographic concentration of high OUD prevalence in Appalachian counties of Kentucky also suggests that rates are likely as elevated in other states across the Appalachian region.

The SAMHSA behavioral health barometer, which is a combination of NSDUH and N-SSATS, reported an OUD prevalence in Kentucky from 2017 to 2019 of 1.3% among persons aged 12 years and older in the past year (Substance Abuse and Mental Health Services Administration, 2020b). We note that the number of people in Medicaid alone with OUD for those aged 18–64 is more than the number estimated by NSDUH data for those ages 12 and over (56,000) via a special tabulation provided to us by SAMHSA prior to our analysis (Substance Abuse and Mental Health Services Administration, 2021). Our OUD prevalence estimate includes only persons aged 18–64 and is higher than the SAMHSA results for the 12 and older population. Even if we included the entire 2019 Kentucky population in the denominator (4,467,673 people) and left the numerator unaltered, the total estimated prevalence of OUD in Kentucky would approach 3.6%, nearly three times higher than the SAMHSA estimate. Further, if we exclude previously unknown cases of OUD and only compare those cases that are directly identified in Kentucky data sets (i.e., known cases), OUD is still three times more prevalent than using NSDUH estimates (Substance Abuse and Mental Health Services Administration, 2020c). A similar underestimation of heroin use prevalence was noted in a recent study comparing NSDUH survey estimates to estimates of heroin use prevalence from other methods (Reuter, Caulkins, and Midgette 2021). Other capture-recapture analyses have reported similar findings and highlight the limitations of surveys for drug use behavior, which is highly stigmatized and criminalized. While this study used four datasets, capture-recapture methods can be used with additional data sources. Using additional data sources in future studies may provide additional information to further improve estimation by either (1) identifying additional persons “known” to have OUD and producing lower estimates of “unknown” persons, or by (2) showing that the number of persons with “unknown” OUD is larger than this analysis estimates leading to increased OUD prevalence estimates. In addition to yielding improved OUD prevalence estimates, additional information could produce more precise (narrower) confidence intervals for clinicians and policymakers alike. The intervals in this manuscript already provide conservative lower bounds for OUD prevalence in each county.

One surprising finding was that Kentucky’s two largest counties had disparate trends in OUD prevalence across the two years: Jefferson County increased and Fayette County decreased substantially. In further analyses, we found that these changes occurred in the same direction as these counties’ opioid-related mortality rates from 2018 to 2019. The extent to which changing mortality rates impacts our estimates is unclear and we cannot quantitatively attribute these single year changes in mortality to any specific intervention. Even so, recent surveillance signals for Kentucky’s overdose mortality (>60% increases in opioid-related deaths in both counties) and EMS runs leads us to expect large increases in OUD prevalence in 2020 due to aspects of the COVID-19 pandemic (Kentucky Injury Prevention & Research Center, 2021; Slavova et al., 2021, 2020). Examining this type of disparate variation provides us with a new opportunity to examine contextual differences in these counties that might be amenable to policy change. Also, this analysis adds credence to the significant value of linked data sets across state agencies to increase information available at the county-level (Barocas et al., 2018).

We employed a novel adaptation of the capture-recapture methodology to arrive at our estimates. Previously, the capture-recapture method had been used to estimate OUD in places with large geographic areas and in which data sets had substantial overlap, such as Massachusetts (Barocas et al., 2018). But Kentucky is a predominately rural state with 120 counties and a total land area that is nearly four times that of Massachusetts, a more densely populated state consisting of only 14 counties. This combination of geographic factors along with fewer available integrated data sets than Massachusetts meant that there was less overlap between data sets and substantially more instances of data sources with no overlapping individuals in given strata. Our approach adapts the methods used in Massachusetts to accommodate fewer data sets and a greater number of counties. The methodologic adaptation from this analysis provides a useful roadmap for other states or jurisdictions that may have a large number of counties and few datasets to include. This will likely be useful in estimating problematic stimulant use, which is increasing in the U.S. but still less common than OUD.

This analysis is not without limitations. The capture-recapture methodology is an imperfect tool for estimation and is bound by several underlying assumptions. First, these estimation methods assume that the identifications from different data sources are independent (Bird and King, 2018; Jones et al., 2020). We utilized data sets in which there may have been dependence among case identification probabilities, and our modeling methods were able to account for this dependence as needed. For example, a person identified in the Medicaid data set may be more likely to also be identified in the PDMP than someone not identified in the Medicaid dataset by virtue of the fact that they have prescriptions paid for by health insurance. In our analyses, the model selection process allowed for inclusion of two-way interaction effects to avoid making the previously described independence assumption. If an interaction effect was selected during the modeling procedure, then independence was not assumed for the corresponding pair of data sets.

Second, the capture-recapture method requires an individual to have non-zero probability of being recorded. In our analysis, Medicaid claims data has only Medicaid enrollees, whereas other data sources might include people outside of Medicaid. Third, our ability to identify OUD variables in each dataset is limited by the available data within datasets and presupposes the validity of administrative data (The Medicaid Outcomes Distributed Research Network (MODRN) et al., 2021). A unified approach to coding and reporting OUD within available data sets is necessary for more accurate estimates. Finally, limitations in the scope of the PDMP caused those data to include only individuals receiving buprenorphine, while methadone and injectable naltrexone are also used as MOUD.

5. Conclusion

In conclusion, our study of OUD prevalence in persons aged 18–64 in Kentucky demonstrates that existing state-level numbers, based on traditional surveys like NSDUH, are likely under-estimates of the prevalence. Furthermore, our estimates indicate year-over-year increases in OUD case counts. Accurate estimates of prevalence are crucial for resource allocation and developing targeted interventions that are appropriately scaled to the size of the problem. Policymakers can use estimates from these analyses to advocate for more resources and put an end to the opioid overdose epidemic in Kentucky.

Supplementary Material

1

Highlights.

  • We employed a capture-recapture method to estimate opioid use disorder prevalence in Kentucky

  • Statewide OUD prevalence was 5.5% and 5.9% for 2018 and 2019, respectively, for adults 18–64 years

  • County prevalence ranged from 1.3% to 17.7%

  • Total estimated prevalence of OUD in Kentucky is nearly three times higher than previous estimates

Acknowledgements

Role of the Funding Source:

This research was supported by the National Institutes of Health and the Substance Abuse and Mental Health Services Admiistration through the NIH HEAL Initiative under award numbers UM1DA049406 and UM1DA049412 with additional support from the National Institute on Drug Abuse [DP2DA051864] to J.A.B. The funder played no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; nor the decision to submit the manuscript for publication. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or its NIH HEAL Initiative. Drs. Chandler and Villani were substantially involved in UM1DA049406 and UM1DA049412, consistent with their role as Scientific Officers. They had no substantial involvement in the other cited grant. The views and opinions expressed in this manuscript are those of the authors only and do not necessarily represent the views, official policy or position of the U.S. Department of Health and Human Services or any of its affiliated institutions or agencies.

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Trial Registration: ClinicalTrials.gov Identifier: NCT04111939

This study protocol (Pro00038088) was approved by Advarra Inc., the HEALing Communities Study single Institutional Review Board (sIRB).

Conflicts of interest:

None of the authors declares any competing or conflicts of interest.

References:

  1. Appalachian Regional Commission, 2022. County Economic Status and Distressed Areas in Appalachia [WWW Document]. URL https://www.arc.gov/appalachian_region/CountyEconomicStatusandDistressedAreasinAppalachia.asp (accessed 5.24.22).
  2. Barocas JA, White LF, Wang J, Walley AY, LaRochelle MR, Bernson D, Land T, Morgan JR, Samet JH, Linas BP, 2018. Estimated Prevalence of Opioid Use Disorder in Massachusetts, 2011–2015: A Capture-Recapture Analysis. American Journal of Public Health 108, 1675–1681. 10.2105/AJPH.2018.304673 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Bird SM, King R, 2018. Multiple Systems Estimation (or Capture-Recapture Estimation) to Inform Public Policy. Annu Rev Stat Appl 5, 95–118. 10.1146/annurev-statistics-031017-100641 [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Centers for Disease Control and Prevention, 2022. Vital Statistics Rapid Release - Provisional Drug Overdose Data [WWW Document]. Vital Statistics Rapid Release. URL https://www.cdc.gov/nchs/nvss/vsrr/drug-overdose-data.htm (accessed 5.24.22). [Google Scholar]
  5. Centers for Disease Control and Prevention, 2021. Drug Overdose Deaths in the U.S. Top 100,000 Annually [WWW Document]. URL https://www.cdc.gov/nchs/pressroom/nchs_press_releases/2021/20211117.htm (accessed 5.24.22).
  6. Chan L, Silverman B, Vincent K, 2019. SparseMSE: “Multiple Systems Estimation for Sparse Capture Data.” R package version 2.0.1 Published online 2019. https://CRAN.R-project.org/package=SparseMSE. [Google Scholar]
  7. Chan L, Silverman BW, Vincent K, 2021. Multiple Systems Estimation for Sparse Capture Data: Inferential Challenges When There Are Nonoverlapping Lists. Journal of the American Statistical Association 116, 1297–1306. 10.1080/01621459.2019.1708748 [DOI] [Google Scholar]
  8. Chen WJ, Fang C-C, Shyu R-S, Lin K-C, 2006. Underreporting of illicit drug use by patients at emergency departments as revealed by two-tiered urinalysis. Addict Behav 31, 2304–2308. 10.1016/j.addbeh.2006.02.015 [DOI] [PubMed] [Google Scholar]
  9. Comiskey CM, Barry JM, 2001. A capture-recapture study of the prevalence and implications of opiate use in Dublin. Eur J Public Health 11, 198–200. 10.1093/eurpub/11.2.198 [DOI] [PubMed] [Google Scholar]
  10. Esri, 2021. ArcGIS Pro 2.8.3 ESRI Inc, Redlands, California. [Google Scholar]
  11. Hay G, Richardson C, 2016. Estimating the Prevalence of Drug Use Using Mark-Recapture Methods. Statistical Science 31, 191–204. 10.1214/16-STS553 [DOI] [Google Scholar]
  12. Hickman M, Cox S, Harvey J, Howes S, Farrell M, Frischer M, Stimson G, Taylor C, Tilling K, 1999. Estimating the prevalence of problem drug use in inner London: a discussion of three capture-recapture studies. Addiction 94, 1653–1662. 10.1046/j.1360-0443.1999.941116534.x [DOI] [PubMed] [Google Scholar]
  13. Hickman M, Taylor C, 2005. Indirect Methods to Estimate Prevalence, in: Sloboda Z (Ed.), Epidemiology of Drug Abuse. Springer US, Boston, MA, pp. 113–131. 10.1007/0-387-24416-6_8 [DOI] [Google Scholar]
  14. International Business Machines Corporation, 2021. InfoSphere Information Server [WWW Document]. URL https://www.ibm.com/analytics/information-server (accessed 5.24.22).
  15. Jones HE, Harris RJ, Downing BC, Pierce M, Millar T, Ades AE, Welton NJ, Presanis AM, De Angelis D, Hickman M, 2020. Estimating the prevalence of problem drug use from drug-related mortality data. Addiction 115, 2393–2404. 10.1111/add.15111 [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Kaiser Family Foundation, 2022. Drug Overdose Death Rate (per 100,000 population) [WWW Document]. KFF. URL https://www.kff.org/other/state-indicator/drug-overdose-death-rate-per-100000-population/ (accessed 5.24.22). [Google Scholar]
  17. Kentucky Injury Prevention & Research Center, 2021. Drug Overdose and Related Comorbidity County Profiles [WWW Document]. URL https://kiprc.uky.edu/programs/overdose-data-action/county-profiles (accessed 5.24.22).
  18. Khazaei S, Poorolajal J, Mahjub H, Esmailnasab N, Mirzaei M, 2012. Estimation of the Frequency of Intravenous Drug Users in Hamadan City, Iran, Using the Capture-recapture Method. Epidemiol Health 34, e2012006. 10.4178/epih/e2012006 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Krawczyk N, Rivera BD, Jent V, Keyes KM, Jones CM, & Cerdá M (2022). Has the treatment gap for opioid use disorder narrowed in the US?: A yearly assessment from 2010 to 2019. International Journal of Drug Policy, 103786. 10.1016/j.drugpo.2022.103786 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Lasher L, Jason Rhodes MPA AC and Viner-Brown S, 2019. Identification and description of non-fatal opioid overdoses using Rhode Island EMS data, 2016–2018. Rhode Island Medical Journal, 102(2), pp.41–45. [PubMed] [Google Scholar]
  21. Mastro TD, Kitayaporn D, Weniger BG, Vanichseni S, Laosunthorn V, Uneklabh T, Uneklabh C, Choopanya K, Limpakarnjanarat K, 1994. Estimating the number of HIV-infected injection drug users in Bangkok: a capture--recapture method. Am J Public Health 84, 1094–1099. 10.2105/AJPH.84.7.1094 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Mattson CL, Tanz LJ, Quinn K, Kariisa M, Patel P, Davis NL, 2021. Trends and Geographic Patterns in Drug and Synthetic Opioid Overdose Deaths — United States, 2013–2019. MMWR Morb. Mortal. Wkly. Rep 70, 202–207. 10.15585/mmwr.mm7006a4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. National Archives, 2016. Soundex System [WWW Document]. National Archives. URL https://www.archives.gov/research/census/soundex (accessed 5.24.22). [Google Scholar]
  24. National Center for Health Statistics, 2021. Bridged-Race Population Estimates, United States July 1st resident population by state, county, age, sex, bridged-race, and Hispanic origin, on CDC WONDER On-line Database. [WWW Document]. URL https://wonder.cdc.gov/wonder/help/bridged-race.html# (accessed 5.24.22).
  25. Nielsen S, Hansen JF, Hay G, Cowan S, Jepsen P, Omland LH, Krarup HB, Søholm J, Lazarus JV, Weis N, Øvrehus A, Christensen PB, 2020. Hepatitis C prevalence in Denmark in 2016—An updated estimate using multiple national registers. PLoS One 15, e0238203. 10.1371/journal.pone.0238203 [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Origer A, 2012. Prevalence of problem drug use and injecting drug use in Luxembourg: a longitudinal and methodological perspective. Eur Addict Res 18, 288–296. 10.1159/000337211 [DOI] [PubMed] [Google Scholar]
  27. Reuter P, Caulkins JP and Midgette G, 2021. Heroin use cannot be measured adequately with a general population survey. Addiction, 116(10), pp.2600–2609. 10.1111/add.15458 [DOI] [PubMed] [Google Scholar]
  28. Seber GAF, 2002. Estimation of animal abundance and related parameters. Blackburn Press, Caldwell, N.J. [Google Scholar]
  29. Slavova S, Quesinberry D, Hargrove S, Rock P, Brancato C, Freeman PR, Walsh SL, 2021. Trends in Drug Overdose Mortality Rates in Kentucky, 2019–2020. JAMA Network Open 4, e2116391–e2116391. 10.1001/jamanetworkopen.2021.16391 [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Slavova S, Rock P, Bush HM, Quesinberry D, Walsh SL, 2020. Signal of increased opioid overdose during COVID-19 from emergency medical services data. Drug and Alcohol Dependence 214, 108176. 10.1016/j.drugalcdep.2020.108176 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Substance Abuse and Mental Health Services Administration, 2021. Special Tabulation using data from the National Survey on Drug Use and Health, 2015–2018. SAMHSA. [Google Scholar]
  32. Substance Abuse and Mental Health Services Administration, 2020a. Key Substance Use and Mental Health Indicators in the United States: Results from the 2020 National Survey on Drug Use and Health. SAMHSA. [Google Scholar]
  33. Substance Abuse and Mental Health Services Administration, 2020b. Behavioral Health Barometer: Kentucky, Volume 6. SAMHSA. [Google Scholar]
  34. Substance Abuse and Mental Health Services Administration, 2020c. 2018–2019 National Survey on Drug Use and Health: Model-Based Prevalence Estimates (50 States and the District of Columbia). SAMHSA. [Google Scholar]
  35. Substance Abuse And Mental Health Services Administration, 2015. National Survey on Drug Use and Health: 4-Year R-DAS (2002 to 2005, 2006 to 2009, and 2010 to 2013): Version 2. 10.3886/ICPSR34415.V2 [DOI]
  36. The Medicaid Outcomes Distributed Research Network (MODRN), Brown E, Schutze M, Taylor A, Jorgenson D, McGuire C, Brown A, Middleton A, Woodcock C, LaPres M, Cohn L, Dowler S, Sandoe E, Rose R, Applegate M, Markman K, Rizzutti M, Truex-Powell E, Ashmead R, Mack A, Bailey E, Kelley D, James AE, Costlow M, Sharbaugh M, Harrell A, Walker L, Becker J, Parsons C, Cai Y, Tyska S, Voskuil K, Donohue JM, Jarlenski MP, Kim JY, Tang L, Ahrens K, Allen L, Austin A, Barnes AJ, Burns M, Chang C-CH, Clark S, Cole E, Crane D, Cunningham P, Idala D, Junker S, Lanier P, Mauk R, McDuffie MJ, Mohamoud S, Pauly N, Sheets L, Talbert J, Zivin K, Gordon AJ, Kennedy S, 2021. Use of Medications for Treatment of Opioid Use Disorder Among US Medicaid Enrollees in 11 States, 2014–2018. JAMA 326, 154. 10.1001/jama.2021.7374 [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Tilling K, Sterne JAC, 1999. Capture-Recapture Models Including Covariate Effects. American Journal of Epidemiology 149, 392–400. 10.1093/oxfordjournals.aje.a009825 [DOI] [PubMed] [Google Scholar]
  38. Wesson P, Lechtenberg R, Reingold A, McFarland W, Murgai N, 2018. Evaluating the Completeness of HIV Surveillance Using Capture-Recapture Models, Alameda County, California. AIDS Behav 22, 2248–2257. 10.1007/s10461-017-1883-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Wood F, Bloor M, Palmer S, 2000. Indirect prevalence estimates of a national drug using population: The use of contact-recontact methods in Wales. null 2, 47–58. 10.1080/136985700111440 [DOI] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHMS1855191-supplement-1.docx (2.6MB, docx)

RESOURCES