Reductions in sustained prescription opioid use within the US between 2017 and 2021

Over the last decade, various efforts have been made to curtail the opioid crisis. The impact of these efforts, since the onset of the COVID-19 pandemic, has not been well characterized. We sought to develop national estimates of the prevalence of sustained prescription opioid use for a time period spanning the COVID-19 pandemic (2017–2021). We used TRICARE claims data (fiscal year 2017–2021) to identify patients who were prescription opioid non-users prior to receipt of a new opioid medication. We evaluated eligible patients for subsequent sustained prescription opioid use. The prevalence of sustained prescription opioid use during 2020–2021 was compared to 2017–2019. We performed multivariable logistic regression analyses to adjust for confounding. We performed secondary analyses that accounted for interactions between the time period and age, as well as a proxy for socioeconomic status. We determined there was a 68% reduction in the odds of sustained prescription opioid use (OR 0.32; 95% CI 0.27, 0.38; p < 0.001) in 2020–2021 as compared to 2017–2019. Significant reductions were identified across all US census divisions and all patient age groups. In both time periods, the plurality of encounters associated with initial receipt of an opioid that culminated in sustained prescription opioid use were associated with non-specific primary diagnoses. We found significant reductions in sustained prescription opioid use in 2020–2021 as compared to 2017–2019. The persistence of prescribing behaviors that result in issue of opioids for poorly characterized conditions remains an area of concern.


Statistical analysis
The primary outcome in this investigation was the prevalence of sustained prescription opioid use, defined as the number of sustained prescription opioid users divided by the number of opioid non-users issued at least one opioid prescription.The primary predictor was the time of initial opioid prescription, with the cutoff established at March 1, 2020.In this context we compared the years 2020-2021, associated with the COVID-19 pandemic to the pre-pandemic time period of 2017-2019.All other abstracted variables were considered covariates in adjusted analyses.Descriptive statistics were calculated for all study variables using the chi-square test for bivariate comparisons.We performed multivariable logistic regression analyses to adjust for confounding from included covariates.In all adjusted analyses, based on a previously validated approach 16 , we accounted for missing race with an imputation method using reweighted estimating equations.Aligned with previous research 7 , we included US census region in adjusted analyses to account for variation in the prevalence of COVID-19 virus and the extent of local/regional government and health department restrictions.We reported all results of regression testing as odds ratios (OR) with 95% confidence intervals (CI) and p-values.We established statistical significance, a-priori, for variables with OR and 95% CI exclusive of 1.0 and p < 0.05.We performed secondary analyses that accounted for interactions between the time period and age, as well as sponsor rank, as our proxy for socioeconomic status.In these comparisons, age 55 and over and the time period 2017-2019 and Senior Officer rank and 2017-2019 were used as the referents.All statistical testing was conducted using SAS v9.4 (SAS Inst., Cary, NC) or STATA v15.1 (STATA Corp., College Station, TX).All methods were carried out in accordance with relevant guidelines and regulations.All conduct of this research and reporting follow the STROBE guidelines.All experimental protocols were approved by our institutional committee at the Uniformed Services University of the Health Sciences prior to commencement.As this was a retrospective review of previously collected de-identified claims-based data, the need for informed consent was waived by the Institutional Review Board of the Uniformed Services University of the Health Sciences.

Results
We identified 1,478,308 individuals who were opioid non-users and issued at least one opioid prescription between 2017 and 2021.Fifty-five percent of the cohort was male with the plurality aged 18-24 (37%) and of White race (45%).The majority of the cohort was composed of civilians (40% dependents, 15% retirees and 0.1% Other) and resided in the South (Table 1).Half the individuals were of Senior Enlisted sponsor rank and 32% were in the Junior Enlisted category.There were relatively marginal differences in the sociodemographic and clinical composition of the cohort treated in 2017-2019 as compared to 2020-2021, although most findings were statistically significant given the size of our sample.
The overall number of opioid prescriptions was significantly reduced over the course of the study period, from 49.5 per 100 individuals in the covered population in 2017-2019 to 10.4 per 100 individuals in the covered population during 2020-2021 (p < 0.001).Between 2017 and 2019, we found that the prevalence of sustained prescription opioid use was 0.87%.The prevalence in the time period 2020-2021 was 0.3%.In adjusted analysis accounting for all confounders, we determined there was a 68% reduction in the odds of sustained prescription opioid use (OR 0.32; 95% CI 0.27, 0.38; p < 0.001) in 2020-2021 as compared to 2017-2019.The prevalence of sustained prescription opioid use by census division ranged from 1.24% in the East South Central to 0.66% in the Pacific region during 2017-2019.Meaningful reductions were appreciated across all census divisions during 2020-2021 (Fig. 1), ranging from 0.54% in the East South Central to 0.19% in the Pacific region.All reductions in sustained prescription opioid use by census division, for 2020-2021 as compared to 2017-2019, were significant (p = 0.001 for New England; p < 0.001 for all other census divisions).
In our secondary analysis assessing for interactions between age and time period, we found significant reductions across all age groups for the period 2020-2021 as compared to individuals 55 and older in 2017-2019 as the referent (Table 2).When evaluating interactions between the time period and our proxy for socioeconomic status (Table 3), as compared to the referent, there was a significant increase in odds of sustained prescription opioid use in individuals of the lowest socioeconomic strata (e.g.Junior Enlisted; OR 1.92; 95% CI 1.51, 2.45; p < 0.001) during 2017-2019 (Table 3).This was reduced during 2020-2021, where no significant difference was appreciated (OR 0.75; 95% CI 0.49, 1.16; p = 0.19).Similarly, the odds of sustained prescription opioid use among Senior Enlisted were significantly increased during 2017-2019 (OR 2.13; 95% CI 1.75, 2.60; p < 0.001).However, during 2020-2021, there were significantly lower odds of sustained prescription opioid use for this subgroup (OR 0.64; 95% CI 0.48, 0.84; p = 0.002).

Discussion
Since the opioid crisis entered the popular consciousness in 2013, numerous efforts at the federal, state and local levels have been implemented to reduce the number of opioid medications circulating in the community and the prevalence of sustained prescription opioid use, non-prescribed opioid use and addiction [3][4][5][17][18][19][20][21][22] . While ome putative success in reducing the number of opioid prescriptions was initially appreciated 3,4,12,13 , the ultimate effectiveness of these various initiatives has not been extensively studied in the context of COVID-19, especially in light of alterations in healthcare delivery that occurred due to the disruptions of the pandemic 6,7,12,20 .This investigation is among the first we are of aware of to consider the prevalence of sustained prescription opioid use in a national sample over a time frame that accounts for the effects of the COVID-19 pandemic.
Overall, we believe that our findings are encouraging as they demonstrate sizable reductions in sustained prescription opioid use in all census divisions, and for all sociodemographic and clinical subgroups considered, from 2017 to 2019 to 2020-2021.In particular, high-risk cohorts as characterized in the literature 9,12,17,19,22 , such as those of younger age and patients from lower socioeconomic strata, experienced significant reductions in the likelihood of sustained prescription opioid use during 2020-2021.Furthermore, it is also encouraging that as compared to the time period 2006-2014 14 , the total number of new sustained prescription opioid users in the Military Health System reduced substantially, from 117,118 in 2006-2014 to 11,648 during 2017-2021.
Nonetheless, the fact that the plurality of individuals in both cohorts who developed sustained prescription opioid use received their initial prescription for poorly characterized conditions, or ailments for which opioid prescriptions are not considered standard of care, is worrisome.These same prescribing behaviors were documented in the Military Health System during 2006-2014, where Other ill-defined conditions and Encounter for administrative purposes were among the most frequent diagnoses associated with the receipt of an opioid prescription in both military and civilian-run facilities 14 .This may stem from the fact that clinical burdens on providers lead them to default to simplified coding practices that do not require clear specificity around the rationale for the opioid medication issued.At the same time, the observed lack of improvement across the 15-year time frame spanning our investigation and that of the earlier study 14 , suggests a lack of efficacy regarding government mandated educational efforts in opioid stewardship and an opportunity for meaningful change going forward.
Vol:.( 1234567890 www.nature.com/scientificreports/Our finding regarding the regional distribution of opioid prescribing behaviors is reasonably well aligned with prior studies on this topic, specifically with higher observed rates of opioid prescribing in the Southern US and the Midwest, as compared to New England and the Northeast 23 .The sociodemographic factors we identified as significantly associated with sustained prescription opioid use, especially during 2017-2019, are also similar to those encountered in previous works, including the influence of socioeconomic status 9,12,17,19,22 .We believe this consistency speaks to the external validity of our findings and their relevance to current health policy 24 .www.nature.com/scientificreports/This relevance is further strengthened by the national scope of our data, development from a very recent clinical cohort with characteristics that allow for generalization to the US demographic aged 64 and younger and the ability to capture the receipt of opioid prescriptions irrespective of residential location or the environment of care 8,10 .Our definition of sustained prescription opioid use also followed established best practices with respect to the use of a claims-based study source 14,15 .As a result, we maintain that our results hold meaning for clinicians, healthcare facilities and policy makers, and that they extend to the federal as well as civilian health sectors 24 .
The overarching findings suggest that the various regulatory initiatives reporting on and restricting the receipt of opioid medications may be making effective headway on reducing the number of opioid non-users who transition to sustained prescription opioid use across the country.Further qualitative research, or mixed methods work, within the Military Health System could potentially identify the practices and behaviors that have led to the reductions in sustained prescription opioid use during 2020-2021.These future findings may be scalable to other healthcare settings nationwide.At the same time, our results call for a narrower focus on the clinical rationale for an opioid prescription, and its alignment with best practices and recommendations regarding care.In this context, we suggest TRICARE and other insurance companies may elect to decline to cover opioid prescriptions that are not associated with an appropriate clinical diagnosis and indication.Additionally, prescribers who routinely issue opioids for inappropriate clinical reasons, or non-indicated clinical conditions, could be identified for additional educational initiatives, clinical practice guidance, or loss of designation as an approved provider.This may be particularly relevant to prescribers within census divisions where the prevalence of sustained prescription opioid use remains higher than the national average, such as the Mountain, East North Central, East South Central and Middle Atlantic states.
There are several limitations inherent to the work which should be recognized.Foremost, this remains a retrospective study using claims-based data, with all the inherent drawbacks associated with such a study design and data source.We do not have access to clinically granular data regarding the decision to issue an opioid prescription, the underlying rationale, or how these relate to the claims-based diagnosis codes reported to TRICARE.Discrepancies in coding practices and surveillance cannot be quantified and remain a potential source of bias that cannot be accounted for.Additionally, our study-specific definitions are predicated on the assumption that patients used opioid medications as directed by the prescribers and we are unable to evaluate non-prescription opioid use, diversion and consumption of illegally obtained prescription opioids, or heroin.This includes, our study specific definition of chronic opioid use, which was based on the work of Oleisky et al. 15 Oleisky et al. maintained that 6-month of sustained use demonstrated good fidelity when defining chronic opioid use using claims-based data 15 .Chronic prescription opioid use has been defined differently in other research 25 and the parameters used to establish the presence of this condition should be noted when making comparisons between studies.Viewed in this light, our estimations would be considered conservative.
Given the nature of our source population, the findings should not be generalized to patients aged 65 and older, those covered by Medicare, or those receiving treatment in VA facilities, as these represent specific societal subpopulations that may not share the characteristics of our sample.Similarly, our results are specific to a population of patients who were previously opioid non-users and cannot be extrapolated to long-term chronic users of opioid medications as a result.Several recent studies have suggested that the characteristics and clinical experience of individuals with longstanding histories of chronic opioid use have worsened in the last 5 years in terms of the inability to discontinue opioids, transition to non-prescription use and use of illegal formulations, as well as episodes of overdose and death 9,17,18,20,22,24 .
In conclusion, this investigation is among the first we are aware of to comprehensively investigate the prevalence of sustained prescription opioid use among previously non-opioid using individuals in the time-period 2017-2021.We are encouraged by our findings which demonstrate temporal improvements as compared to historical reports and significant reductions in sustained prescription opioid use in 2020-2021 as compared to 2017-2019.The persistence of prescribing behaviors that result in issue of opioids for poorly characterized

Figure 1 .
Figure 1.The prevalence of sustained prescription opioid use by census division in the United States from 2017 to 2019 (Pre-pandemic) compared to 2020-2021 (Pandemic).

Table 1 .
Baseline sociodemographic and clinical characteristics of the study cohort.

Table 2 .
Results of the multivariable logistic regression analysis regarding factors associated with the development of sustained prescription opioid use, accounting for interactions between the time period and patient age at time of receipt of initial opioid prescription.

Table 3 .
Results of the multivariable logistic regression analysis regarding factors associated with the development of sustained prescription opioid use, accounting for interactions between the time period and sponsor rank, our proxy for socioeconomic status.