Our analyses suggest an unprecedented concentration of opioid use among enrollees with CNCP diagnoses who have used some opioids. In HealthCore enrollees, the top 5% of users accounted for over 70% of total opioids used in 2005. The top 1% accounted for 43% of total opioids used. Between 2000 and 2005, opioid use became more concentrated in these groups. In Arkansas Medicaid enrollees, the top 5% of users accounted for 47% of total opioids used and the top 1% accounted for 21% of total opioids used. This concentration of use is not seen with any other prescribed medication and raises unique challenges and opportunities for reducing the risks of chronic opioid use. The use of opioids was more heavily concentrated in the top 1% of users in HealthCore (as defined by total morphine equivalents of opioid use in the year) than in the top 1% of Arkansas Medicaid users not because the top 1% in HealthCore were heavier utilizers than the top 1% in Arkansas Medicaid. Rather, the top 1% in HealthCore had a higher mean daily dose than those in Arkansas Medicaid but a lower days supply, however both had days of supply >365 days indicating the use of multiple concurrent opioids. The mean user in Arkansas Medicaid actually had heavier use as defined by total morphine equivalents received in the year than the mean user in HealthCore. Indeed, individuals in the top 1% of use in Arkansas Medicaid utilized on mean 88,275 mg of morphine equivalents in year 2005, while individuals in the top 1% of HealthCore utilized on mean 55,800 mg of morphine equivalents.
In both groups, the likelihood of heavy opioid use increased with the number of CNCP diagnoses. Only in the HealthCore group did heavy use increase with increasing number of mental health and substance abuse diagnoses. The higher prevalence of mental health and substance abuse comorbidity in the Medicaid sample may account for less concentration of opioid use in this group. High opioid utilization was associated with age 41-60 in both samples and also age >60 in HealthCore. High utilization was associated with all the CNCP diagnoses and substance abuse disorders in both samples.
Our results suggest that among individuals in the HealthCore sample with CNCP diagnoses who receive opioids, most are not chronic users of opioids. For example, in HealthCore in 2005, it appears that between 10% and 20% percent of the sample were chronic users, as the 80th percentile for days supply was 43 days, which is clearly not chronic use, while the 90th percentile for days supply was 150 days, which might be considered chronic use. On the other hand, in Arkansas Medicaid, chronic use was more common, with 30% of the Arkansas Medicaid sample utilizing 136 days or more.
Our study suggests that mean daily dose did not change between 2000 and 2005. A current controversy in pain management is whether there should be any upper limit or cautionary range for daily opioid dose. The Washington State AMDG opioid dosing guidelines for primary care physicians suggest that “rarely, and only after pain management consultation, should the total daily dose of opioid be increased above 120 milligram oral morphine equivalents” 31 While our study was not designed to investigate what constitutes an appropriate upper limit of daily opioid use, we can assess what percentage of individuals with CNCP disorders and opioid use exceed given thresholds. For example, in HealthCore, in both 2000 and 2005 about 8% of individuals with CNCP who were on opioids had mean daily doses greater than 120 mg oral morphine equivalents, and in Arkansas Medicaid about 5% of individuals exceeded this threshold.
Our findings suggest several key areas for future research. First, given that opioid use is so heavily concentrated among relatively few users, are the negative outcomes from opioids similarly concentrated among relatively few users, and are these groups the same? If this is the case, efforts to minimize the negative outcomes of opioid use might have to focus only on a relatively small percentage of opioid users. We also need to better understand the characteristics of heavy utilizers. While our results suggest that heavy opioid utilization is associated with more CNCP and MH/SUD diagnoses, clearly additional research is needed to define the characteristics of these individuals, and this might need to be done for different sub-populations. For example, in Healthcore, the elderly were more likely to be heavy utilizers, and there were no gender differences. On the other hand, in Arkansas Medicaid, middle age individuals and men were more likely to be heavy utilizers. Also, there may be other important factors associated with high utilization of opioids that our data do not allow us to address. For example, work (or disability) status, social support, and physical health status might all influence use of opioids, and this should be pursued in future studies.
We need to better understand the patterns of utilization of these individuals. We speculate that there may be distinct, clinically important patterns of use, and that some patterns may be more suggestive of misuse than other patterns. For example, some high utilizers may be patients on high, but stable, doses of opioids, prescribed by one clinician. These individuals might be on two regularly prescribed opioids, such as long acting opioid for control of baseline pain, and a short acting opioid for breakthrough pain, but with no early re-fills. Other high utilizers may have pharmaco-epidemiological profiles which are characterized by increasing dosages, early re-fills, and prescriptions from multiple clinicians and pharmacies—all patterns suggestive of misuse, or possibly pseudo-addiction. We believe a key step in this line of research is determining what percentage of individuals have pharmacy profiles that fit into the first category, and what percentage have pharmacy profiles that are consistent with the second category, as this will give us valuable insight into the magnitude of the possible abuse problem. We need to better understand whether these pharmacy profiles are actually associated with better or worse outcomes. For those individuals whose pharmacy use patterns suggest abuse or misuse, we need to know whether pseudo-addiction associated with poor pain control might be responsible for the observed patterns.
Finally, we need to better understand the pharmaco-epidemiology of opioid use from the vantage point of the clinician. In particular, it would be helpful to know to what extent opioid prescribing is concentrated among a relatively small percentage of physicians. Obviously pain specialists will have high rates of opioid prescribing, but among primary care physicians, is opioid prescribing highly concentrated? Our results are helpful to some degree in this regard, as they allow individual clinicians to compare their own opioid prescribing patterns with the distributions of days supply and mean daily dose in two large, diverse populations. Days supply showed more variation than daily dose and might be considered as a target for harm reduction efforts.
A primary goal of this report was to determine whether the observed increases in opioid use in the TROUP populations between 2000 and 2005 were attributable to high utilizers of opioids. In Arkansas Medicaid, we found that the increased use (total annual milligrams of morphine equivalents) occurred over all percentiles, suggesting that the observed increases were not attributable to high utilizers of opioids. On the other hand, in HealthCore, we found that opioid use increased generally across all percentiles, but the greatest increases occurred among the heaviest users who were older and likely to have multiple chronic pain conditions.
Our results should be interpreted in light of the following limitations. Our data are from diverse sources, but are not nationally representative, so the generalizability to the larger population is unknown. Arkansas Medicaid serves a low income, vulnerable population in the South-Central U.S. HealthCore has enrollees in many states in the Midwest, West, and Southeast, most of whom are working, middle, or upper class. Our intent in investigating these two dissimilar populations was not to draw contrasts, but to describe the current range of opioid prescribing practices. Indeed, the much higher burden of disease in Arkansas Medicaid, measured in terms of both rates of CNPC and MH/SUD diagnoses, makes comparisons problematic.
Because the samples were dissimilar in virtually every way (geographically, sociodemographically, burden of illness) we cannot say what the observed differences in opioid utilization between Healthcore and Arkansas Medicaid were due to. Further, we know of no system factors such as re-imbursement policies or insurance restrictions that might have led to differences among these groups. HealthCore is primarily fee-for-service, and Arkansas Medicaid is entirely so. Co-pays and deductibles are generally be less in Arkansas Medicaid than in HealthCore, but Arkansas Medicaid recipients on average have fewer resources. Further, Arkansas Medicaid recipients are limited to six prescriptions per month.
Although we relied upon conversion factors from published sources to derive morphine equivalents (27
) there are no canonical conversion tables, and estimates of conversion factors differ, generally by small amounts. Differences between conversion tables were resolved by consensus among the clinicians on this study, in collaboration with other researchers and clinicians.
Methadone presents two challenges for a study such as ours. First, published estimates of conversion factors for methadone to morphine equivalents differ considerably (more so than conversion factors for other opioids to morphine equivalents). Second, we were not able to separate methadone used for pain from methadone used for methadone maintenance. However, methadone accounted for a relatively small percentage of total opioid use in our samples, and methadone maintenance is not common. For example, there is only one methadone clinic in all of Arkansas.
We relied upon administrative data for diagnoses and for pharmacy records. No independent clinical assessment of patients to confirm diagnoses could be done. Pain diagnoses have high specificity, although sensitivity is likely lower (33
). Our study does not reflect opioids paid for out of pocket or those bought over the internet without a prescription (34
) or bought illegally or diverting them. As we analyzed each year's data separately and did not track individual subjects’ status from year to year our results represent population trends and not the trends of individual enrollees.
In conclusion, opioid use, as measured by days supply and cumulative yearly opioid dose, is heavily concentrated among a small percent of users. It is increasing broadly across all types of users of opioids, and in the HealthCore commercially insured population, the largest increases were generally among those individuals who are already the heaviest users. The characteristics of these high utilizers need to be further established, and the benefits and risks of their treatment evaluated.