Age and Sex Trends in Long-term Opioid Use in Two Large American Health Systems Between 2000 and 2005

Abstract

Objective

To estimate recent age- and sex-specific changes in long-term opioid prescription among patients with chronic pain in two large American Health Systems.

Design

Analysis of administrative pharmacy data to calculate changes in prevalence of long-term opioid prescription (90 days or more during a calendar year) from 2000 to 2005, within groups based on sex and age (18–44, 45–64, and 65 years and older). Separate analyses were conducted for patients with and without a diagnosis of a mood disorder or anxiety disorder. Changes in mean dose between 2000 and 2005 were estimated, as were changes in the rate of prescription for different opioid types (short-acting, long-acting, and non-Schedule 2).

Patients

Enrollees in HealthCore (N = 2,716,163 in 2000) and Arkansas Medicaid (N = 115,914 in 2000).

Results

Within each of the age and sex groups, less than 10% of patients with a chronic pain diagnosis in HealthCore, and less than 33% in Arkansas Medicaid, received long-term opioid prescriptions. All age, sex, and anxiety/depression groups showed similar and statistically significant increases in long-term opioid prescription between 2000 and 2005 (35–50% increase). Per-patient daily doses did not increase.

Conclusions

No one group showed especially large increases in long-term opioid prescriptions between 2000 and 2005. These results argue against a recent epidemic of opioid prescribing. These trends may result from increased attention to pain in clinical settings, policy or economic changes, or provider and patient openness to opioid therapy. The risks and benefits to patients of these changes are not yet established.

Introduction

Various studies over the last decade have observed increases in the prescription of opioid medications. A study of United States primary care physicians from 1992 until 2001 found a “decade-long increase in opioid prescription.” [1] In fee-for-Service Medicaid between 1996 and 2002, opioid prescriptions were found to have increased at almost twice the rate of non-opioid medications [2–5]. Australia was found to have a “dramatic increase in opioid prescribing” from 1990 to 1996 [4], Nordic countries to have experienced an increase from 2002 to 2006 [6], and Israel an increase from 2000 to 2004 [7]. A Centers for Disease Control report concluded that there has been over the last 10 years “a dramatic increase in the prescription of major types of opioid analgesics, as physicians were encouraged to prescribe stronger analgesics (i.e., opioids) for painmanagement.” [8] Even though the absolute prevalence of regular opioid use among the US population is about 2% [9], these results imply a recent epidemic of opioid prescription for chronic pain.

Different groups of patients may have experienced different increases in use of prescription opioids. The benefits and risks of opioid analgesics vary by age and sex [10,11], and different rates of opioid use have been observed in these groups [12–14]. Older age groups in some studies have been found to be the most prevalent users of opioids [9], and mental health conditions are strongly associated with increased rates of opioid use [15]. Several studies have shown different changes across time in opioid use in groups defined by sex and age [1,16–20], but only in the context of specific diagnoses (such as back pain or terminal cancer), specific geographical regions, specific medications, specific age ranges, substance abuse, or drug-related fatalities. Prior research has not examined generally the trends in the prevalence of long-term opioid prescription in different age and sex groups.

This study explores these trends between 2000 and 2005 in two large United States samples: a national commercially insured health network and a state Medicaid program. The findings from this analysis help to clarify the epidemiology of opioid use in the United States, suggest potential causes for changes in use across time, identify groups with potential risk, inform policy around analgesic prescription, and supplement public health efforts around safe prescribing and pain management.

Methods

Study Populations

Administrative claims data from two populations with different sociodemographic characteristics were analyzed: HealthCore and Arkansas Medicaid. De-identified datasets from 2000 until 2005 were used [2].

HealthCore, Inc., a subsidiary of WellPoint, Inc. has an integrated research database with pharmacy and administrative claims data collected from 14 WellPoint health plans in the U.S. consisting of more than 31 million members. The data were aggregated from five plans in the West, Midwest, and Southeast regions. All plan members were fully insured for medical and pharmacy coverage through commercial insurance products. This population represents a geographically diverse, largely middle-class group of employed individuals and their dependents.

Arkansas Medicaid provides health services to individuals who are eligible because of health status, financial need, or both. Roughly one-third are low-income children, roughly one-sixth qualify for supplemental security income, and roughly 5% receive aid to the Aged, Blind, and Disabled; less than 1% are medically needy [21]. In 2008, about 40% of all vendor payments from Arkansas Medicaid applied to patients aged 21–64, and about 20% to patients 65 and older. Arkansas Medicaid collects claims data from all federally mandated services and almost all federal optional services, including prescription drugs. Over 90% of the enrollees participated in a primary care program, in which primary care physicians coordinated their care. The Arkansas Medicaid program has a prescription benefit limit of three prescriptions per month, with the availability of an extension of prescription benefits up to six per month for most adult beneficiaries. Exceptions to this were persons under 21 years of age and non-dual Long Term Care eligible, who received unlimited prescriptions. Patient co-payments of up to $3 were required for most medications and visits, depending on recipient eligibility category and drug product. Arkansas is one of the poorest states in America, and this Medicaid population constitutes a disadvantaged group. Arkansas also lies in a region of high opioid use and abuse [18], and in 2006 had the second-highest prevalence in the United States of non-medical use of pain relievers (6.4% of those over age 18 [22]).

Analysis

The primary measure in this analysis was long-term opioid use, defined as 90 days or more of prescribed opioid within a calendar year, among patients with a pain-related diagnosis. Tracer chronic pain-related diagnoses were established from claims data as having a claim for one or more of the most common chronic pain conditions: back pain, neck pain, arthritis/joint pain, headache/migraine, and HIV/AIDS (see [2] for more detail on included diagnoses). Patients were categorized as having none vs one or more chronic pain-related diagnoses during each year. Patients with a cancer diagnosis were excluded because the primary interest was in chronic opioid prescription for non-cancer pain.

Patients were grouped by sex and age (18–44, 45–64, and 65 years and older). In order to identify long-term users, the percentage of each group receiving 90 or more days supply in each year from 2000 to 2005 was computed by summing the total number of days of supply for each opioid prescription dispensed during the calendar year. This was estimated by aggregating pharmacy claims by drug type, and adding up the total number of days of which an opioid medication was prescribed. Days where there were overlapping prescriptions of two opioids were counted as part of a single day [2].

All estimates were calculated separately for HealthCore and Arkansas Medicaid. Annual prevalence estimates were graphed and examined for linearity and trends. Correlation coefficients between study year and prevalence were calculated to determine linearity. The linear trend over the 6 years of the study was estimated by fitting a regression line to the yearly prevalence estimates, weighted by the number of observations for each year. The change between 2000 and 2005 was computed as the percentage change in prevalence relative to the baseline prevalence using the estimates from the regression parameters, namely (Predicted Prevalence in 2005—Predicted Prevalence in 2000)/Predicted Prevalence in 2000. The regression parameters were used to estimate change over time because there were some nonlinear yearly variations, and using only data from 2000 and 2005 would omit data from the intervening years. In order to account for the differences in baseline rates, these results describe relative rather than absolute percentage changes. For instance, an increase of prevalence from 5% of the population in 2000 to 10% of the population in 2005 is calculated as a 100% increase, rather than a 5% change.

To compare the percentage change over time between groups, we constructed confidence intervals around the estimates. Because there is no well-established algorithm to establish confidence intervals for relative percentage change over time [23], we applied resampling (bootstrap) techniques to obtain confidence intervals. Resampling methods have been shown to yield reliable results for inference estimates for fractional changes [24]. For each iteration, we randomly selected with replacement the total number of cases found in the sample for each age and sex group, and calculated the observed percentage change using the equation above. One thousand iterations were performed for each group, from which the mean percentage change and 95% confidence intervals were constructed. Confidence intervals that did not overlap zero were interpreted as a statistically significant (at α = 0.05, two-tailed) increase in prevalence over time. To compare change between pairs of groups (different age groups and male compared with female), we calculated the mean and 95% confidence interval for the difference of the estimates. Confidence intervals that did not overlap zero were interpreted as a statistically significant difference (at α = 0.05, two-tailed) in the change between two groups over time.

Subgroup analyses were conducted for patients with and without diagnoses of mood disorders or anxiety disorders. Mood and anxiety disorders were identified and classified from diagnostic codes, using the Agency for Healthcare Research and Quality Clinical Classification Software (AHRQ-CCS). Mood disorders included depressive and bipolar conditions, and anxiety disorders included panic disorder, generalized anxiety disorder, social phobia, and obsessive-compulsive disorder (see [25] for a complete list). Changes in dose over time were estimated by calculating the mean daily dose in morphine equivalents for patients receiving long-term opioids, using established conversion factors. Opioid categories were defined as long-acting DEA Schedule II, short-acting DEA Schedule II, and DEA Schedule III-IV (“non-Schedule II”) (see Table 1 in [2] for a listing of all the opioids by type). Changes in the percentages of patients receiving each of these types of opioid between 2000 and 2005 were calculated.

The re-sampling algorithm was programmed using the programming language Perl (ActiveState Perl); the remaining analyses were conducted in SPSS (Version 10.0).

Results

In HealthCore, data were available for 2,716,163 patients in 2000 and 3,768,223 in 2005. A total of 17.9% had a tracer pain diagnosis in 2000 and 23.8% in 2005. Of those with a tracer pain diagnosis, 57% were female, and the mean age was 45.1 years in 2000 and 45.8 years in 2005. In Arkansas Medicaid, there were 115,914 patients in 2000 and 127,866 in 2005. A total of 31.3% had a tracer pain diagnosis in 2000 and 34.0% in 2005; 75% of those with a tracer pain diagnosis were female, and the mean age was 54.8 years in 2000 and 50.8 years in 2005.

Linear Trends

The analysis of trends showed a linear increase in chronic opioid use for all age and sex groups for each calendar year between 2000 and 2005, in both Arkansas Medicaid and HealthCore, and in those with and without tracer chronic non-cancer pain diagnoses. The smallest correlation coefficient across all age and sex groups for the association between study year and percentage receiving opioids was 0.86, suggesting a consistent linear increase over time in all groups.

Change in Prevalence

Tables 1 and 2 show the prevalence and percentage change in prevalence of long-term opioid use from 2000 to 2005 among patients with a chronic pain diagnosis in HealthCore and Arkansas Medicaid, organized by age and sex. Figure 1 shows the percentage change in prevalence in these groups between 2000 and 2005. In Health-Core, among those with a tracer chronic pain-related diagnosis, in 2000 the highest prevalence of long-term opioid use by age was among men aged 65 and older (4.6%), and women aged 65 and older (6.9%). Between 2000 and 2005 there was an increase in long-term opioid use for all age groups and both sexes. None of the confidence intervals overlapped zero, suggesting a significant increase in all groups during this time. Women aged 18–44 had a significantly lower increase over time (24.1%) than did women aged 45–64 (34.4%) and men aged 45–64 (40.9%), but other comparisons between age groups and between sex groups were not significant at α = 0.05.

Table 1.

Long-term opioid use in HealthCore among patients with a chronic pain diagnosis: Prevalence in 2000 and 2005, and estimated percentage change in prevalence from 2000 to 2005

Group	% with Long-Term Use (N with Pain-Related Diagnosis)		Estimated % Change in Prevalence from 2000 to 2005 (95% Confidence Interval)
	2000	2005
Female, 18–44	2.7 (133,466)	3.5 (227,427)	24.1 (19.2–29.0%)
Female, 45–64	4.4 (120,999)	6.1 (247,469)	34.4 (30.3–38.5%)
Female, 65+	6.9 (23,227)	9.3 (38,529)	35.3 (27.7–42.8%)
Male, 18–44	2.1 (107,610)	2.9 (183,230)	33.8 (27.3–40.4%)
Male, 45–64	3.8 (86,388)	5.4 (174,646)	40.9 (35.3–46.4%)
Male, 65+	4.6 (14,104)	5.9 (26,236)	25.0 (13.7–36.4%)

Table 2.

Long-term opioid use in Arkansas Medicaid among patients with a chronic pain diagnosis: Prevalence in 2000 and 2005, and estimated percentage change in prevalence from 2000 to 2005

Group	% with Long-Term Use (N with Pain-Related Diagnosis)		Estimated % Change in Prevalence from 2000 to 2005 (95% Confidence Interval)
	2000	2005
Female, 18–44	12.1 (9,158)	19.3 (12,655)	53.5 (43.2–63.9%)
Female, 45–64	19.6 (7,616)	29.9 (10,297)	52.1 (43.9–60.3%)
Female, 65+	12.9 (10,861)	18.6 (9,271)	40.2 (30.8–49.7%)
Male, 18–44	14.3 (3,850)	19.8 (5,306)	33.9 (20.8–47.0%)
Male, 45–64	21.4 (2,696)	32.9 (4,114)	52.2 (39.4–65.0%)
Male, 65+	12.4 (2,093)	17.9 (1,872)	40.6 (18.6–62.6%)

Figure 1.

Percentage change in prevalence of long-term opioid use among patients with one or more tracer chronic pain diagnoses from 2000 to 2005, by age and sex group, in Health-Core and Arkansas Medicaid. Vertical bars represent 95% confidence intervals.

In Arkansas Medicaid, between 2000 and 2005 there was an increase in long-term opioid use for all age groups and both sexes. None of the confidence intervals for the change overlapped zero, suggesting there was a statistically significant increase in all groups. Despite some differences between groups in mean change over time, none of the age or sex groups was significantly different from any of the others at α = 0.05.

Mood and Anxiety Disorders

Tables 3 and 4 show the prevalence in 2000 and 2005 of long-term opioid prescription among patients with and without mood disorders and anxiety disorders. The prevalence in patients with mood disorders was substantially higher than that in patients without mood disorders, and similarly for patients with anxiety disorders compared with those without anxiety disorders. Among all the four groups (with or without mood disorder; and with or without anxiety disorder) in both HealthCore and Arkansas Medicaid, none of the confidence intervals for percentage change overlapped zero, with the exception of men aged 65 years and over who had either mood disorder or anxiety disorder. This trend held for both anxiety and depression, and in both Arkansas Medicaid and Health-Core. In all the age- and sex-matched comparisons defined by the presence or absence of a mood or an anxiety disorder, there were no statistically significant differences observed between groups. These results suggest that, despite differences in prevalence, similar increases in prescription of long-term opioids occurred between 2000 and 2005 regardless of the presence of mood and anxiety disorders.

Table 3.

Prevalence of long-term opioid use in HealthCore and Arkansas Medicaid in 2000 and 2005, by presence or absence of mood disorder

Group	HealthCore (% with long-term opioid use)				Arkansas Medicaid (% with long-term opioid use)
	With Mood Disorder		Without Mood Disorder		With Mood Disorder		Without Mood Disorder
	2000	2005	2000	2005	2000	2005	2000	2005
Female, 18–44	7.8	9.4	2.3	2.8	15.1	26.3	2.3	2.8
Female, 45–64	10.6	14.8	4.0	5.1	23.4	35.6	4.0	5.1
Female, 65+	12.6	17.9	6.6	8.5	18.4	30.6	6.6	8.5
Male, 18–44	7.6	9.4	1.9	2.5	18.3	26.0	1.9	2.5
Male, 45–64	11.4	14.6	3.5	4.9	26.8	42.6	3.5	4.9
Male, 65+	8.8	12.9*	4.5	5.5	14.6	27.1*	4.5	5.5

^*Confidence interval (CI) of percentage change between 2000 and 2005 overlaps zero; all other CIs of percentage change do not overlap zero.

Table 4.

Prevalence of long-term opioid use in HealthCore and Arkansas Medicaid in 2000 and 2005, by presence or absence of anxiety disorder

Group	HealthCore (% with long-term opioid use)				Arkansas Medicaid (% with long-term opioid use)
	With Anxiety Disorder		Without Anxiety Disorder		With Anxiety Disorder		Without Anxiety Disorder
	2000	2005	2000	2005	2000	2005	2000	2005
Female, 18–44	7.5	9.2	2.5	3.0	20.7	31.5	10.5	14.8
Female, 45–64	8.6	13.3	4.3	5.6	26.8	41.9	18.6	26.6
Female, 65+	10.2	15.9	6.8	8.8	20.4	31.1	12.6	17.9
Male, 18–44	6.0	8.7	2.0	2.6	23.6	31.1	13.0	17.2
Male, 45–64	9.8	12.8	3.6	5.1	30.7	52.8	20.5	29.4
Male, 65+	10.8	12.4*	4.5	5.7	24.0	35.6*	12.0	17.1

^*Confidence interval (CI) of percentage change between 2000 and 2005 overlaps zero; all other CIs of percentage change do not overlap zero.

Mean Dosage

Table 5 represents the mean daily dosage in morphine equivalents for patients who received opioids, in the years 2000 to 2005. In most of the age and sex groups, there was a slight decrease in mean dose in most of the groups, but all of the confidence intervals overlapped zero, suggesting that the change was not statistically significant in any of the groups, or between the groups.

Table 5.

Mean daily doses (in morphine equivalents) of opioids in HealthCore and Arkansas Medicaid in 2000 and 2005, and estimated % change from 2000 to 2005

	HealthCore			Arkansas Medicaid
	2000	2005	Estimated % Change	2000	2005	Estimated % Change
Female, 18–44	54.2	53.8	0.7	50.4	45.6	−8.0
Female, 45–64	56.1	55.1	−2.1	54.8	52.4	−4.9
Female, 65+	56.5	54.4	−3.8	60.4	54.6	−9.6
Male, 18–44	50.0	50.0	0.9	49.6	48.3	−3.0
Male, 45–64	51.8	52.3	1.8	53.6	53.3	0.5
Male, 65+	53.7	52.1	−2.4	58.3	54.8	−3.8

Type of Opioid

Table 6 shows the percentage changes in different types of opioid prescription between 2000 and 2005, grouped by age and sex. In both databases, non-Schedule 2 opioids had the smallest increases. In all age and sex groups in Health Core, increases in short-acting opioids were greater (70–85%) than those in long-acting opioids (25–50%). In Arkansas Medicaid, the increases were similar between long- and short-acting opioids, except for among men aged 45–64 (115% increase in long-acting, compared with 60.6% for short-acting) and 65 years and over (80.3% increase in long-acting, 60.6% increase in short-acting).

Table 6.

Percentage change in types of opioid prescription from 2000 to 2005 in HealthCore and Arkansas Medicaid

	HealthCore (%)			Arkansas Medicaid (%)
	Long	Short	Non-Schedule 2	Long	Short	Non-Schedule 2
Female, 18–44	30.6	87.0	17.9	35.1	37.6	11.1
Female, 45–64	50.7	80.9	14.3	63.4	48.4	14.8
Female, 65+	50.0	79.3	13.8	56.0	57.2	18.8
Male, 18–44	25.8	83.3	21.3	35.3	26.8	12.4
Male, 45–64	48.0	70.4	13.6	115.5	60.6	11.8
Male, 65+	29.7	69.0	12.6	80.3	50.0	17.9

Discussion

Increasing opioid prescription in the United States has been observed in various populations over the last decade, but has not been investigated by age and sex categories in large datasets. This study found that in two large, dissimilar patient populations, enrollees with a chronic pain diagnosis in all age and sex categories experienced a significant increase of 35–50% in the prevalence of long-term opioid use between 2000 and 2005. The uniformity of this finding, using relatively large samples and two disparate populations, and in groups with higher or lower prevalence, suggests that the trend was generalized, and not isolated to or markedly more pronounced in any subset of the study populations. Men and women of the same age groups did not manifest significantly different changes in long-term opioid prescription over this time period, nor were there significant differences between patients based on the presence of a mood disorder or anxiety disorder. Previous reports from NHANES data have suggested that opioid increases during the 1990s have been concentrated in women, especially those over age 45, but we did not observe this result [26].

The absolute prevalence of opioid prescription in these populations was not high, nor were the increases dramatically large. In HealthCore in 2005, in the group with the highest prevalence (women aged 65 years and older), 9.3% of patients with a chronic pain diagnosis received long-term opioid prescription. In Arkansas Medicaid, men aged 45–64 with a chronic pain diagnosis had the highest prevalence at 32.9%. The largest increases in prevalence were about 50% over the course of 5 years. Put another way, among patients with chronic pain diagnoses, the vast majority did not receive long-term opioid prescriptions. Despite statistically significant increases over time in all the groups observed, and particularly among short-acting opioids, none of the rates reached a level that could be considered epidemic. Given that the main risks associated with prescription opioid use are related to total dose and duration [11], these findings suggest that most patients with chronic pain may not be subject to high levels of risk from chronic opioid therapy. Moreover, the mean daily dosage of opioids among those with long-term use slightly declined in most of the groups during the study period (Table 5), although this decline was not statistically significant. Together these findings suggest that while a greater percentage of patients in the population may have been at risk for harmful consequences from opioids in 2000 than 2005, individual long-term users may not have experienced greater risks from opioid use. Additional research is needed to ascertain how various degrees and patterns of opioid use are associated with harm among different patient populations, especially given recent increases in adverse events related to them [8].

There were striking differences in prevalence of long-term opioid prescription between the two datasets, with Arkansas Medicaid showing two- to tenfold higher prevalence rates than HealthCore for each age group during all the years of the study. This is consistent with other epidemio-logical research that has identified a high rate of opioid use in Arkansas [22], and among individuals with low socioeconomic status [9]. Despite the difference in baseline rates, the relative percentage changes from 2000 to 2005 were similar across most of the groups in Arkansas Medicaid and HealthCore. This similarity in increase is noteworthy, given the two samples’ divergent socio-demographic mix, rates of chronic illness and disability, and baseline rates of long-term opioid prescription. Overall, these findings argue for a diffuse, cultural, or systematic process driving changes in opioid prescription, rather than factors specific to one population, region, sex, or age group.

There is no clear reason for a uniformly increasing prescription of opioids across age and sex groups during this time, but we may speculate. Attention to pain as a “fifth vital sign,” which began in the mid- to late 1990s [27,28], may have contributed to the increases in pain-related diagnoses as well as to prescription of opioids for pain among those with pain. It is still unclear how much the “fifth vital sign” movement changed the prescription of analgesics, as its application has not been found to improve pain management in clinical practice [29]. Patient awareness of or demand for treatments for pain, or for specific opioid analgesics, may have increased during the last decade, although opioids were not typically marketed directly to consumers. Non-medical use of opioids, especially driven by abuse and dependence, has increased recently [22], and patients may have started asking physicians more for prescription opioids for purposes other than pain. While abuse and misuse may account for some of the observed increase, this is not consistent with the general increase observed in all age and sex groups, especially because opioid misuse is observed infrequently in older adults [30–32]. Our analysis found no evidence that patients with mental health conditions such as depression and anxiety accounted for the increase, as similar increases were seen in patients without these diagnoses.

Political or economic forces may have contributed to a uniform increase in long-term opioid prescriptions. During the last 10 years, many states passed intractable pain acts, which legitimized the use of long-term opioid therapy for chronic pain, and may have led to globally increased opioid prescription [33,34]. The current analysis was not able to identify if legislation preceded increases in opioid prescription in the geographical areas in question. Marketing forces such as detailing to physicians may also have made physicians more likely to consider and prescribe opioids, although we know of no specific evidence of increased marketing or detailing during this period. In Italy, opioid consumption was posited to increase starting in 2000 as a result of heavy marketing of newer analgesic agents [35], although novel opioid drugs were not being introduced into the American market during this time period. At the same time, it is difficult to reconcile the observed increase in prescription of opioids with increases in prescription monitoring programs by states, and with increased concern expressed by state medical board members about prosecution and criminal investigation around opioids [36], both of which might naturally lead to less opioid prescriptions.

Prescribers choose between opioid and non-opioid medications to treat pain. Use of non-opioid analgesics may have decreased during the study period, especially with the withdrawal of the widely used COX-2 inhibitors: Rofecoxib (Vioxx) was withdrawn in 2004 and valdecoxib (Bextra) in 2005, and prescriptions of celecoxib (Celebrex) waned due to concerns related to a class effect. Prescribers may have turned to opioid analgesics instead, although our research was not able to ascertain if this effect occurred. More generally, use of all prescription medications in the United States has increased over the last decade [37], although our analysis was not able to identify if opioids showed a greater increase than other classes of medication.

The prevalence estimates and increases observed among those aged 65 years and older are noteworthy. In HealthCore, which represents a large number of geographically diverse, mainly middle-class individuals and their dependents, the highest prevalence of opioid prescription during the study period was among older adults, especially older women: in 2005, 9% of older women with a chronic pain diagnosis received long-term opioid prescription. A similar age-related difference in prevalence has been observed in other studies [9]. The increases in adults aged 65 years and older between 2000 and 2005 in HealthCore and Arkansas Medicaid were substantial, and similar to those in younger populations. Given the potential hazards of opioid use in older adults, such as delirium, respiratory depression, falls, and cognitive impairment, further attention is merited to understand prescription patterns in the groups that are most at risk for adverse side effects. In particular, the safety of low-dose or high-dose opioids for older adults has not been well established; Table 5 indicates that the mean total dose in morphine equivalents was not substantially different between older and younger adults, but did not change much over time, and Table 6 implies that increases happened in prescriptions of both long-acting and short-acting opioids. Additional research examining dosing, outcomes, and safety of opioids in older adults is clearly warranted.

There are several important limitations to this analysis. Using administrative data might misrepresent either medication use or diagnoses, mainly by missing cases. This is a particular concern for diagnoses of Medicare dual eligibles, where Medicare is most often the primary payer. It was estimated that over 90% of prescriptions were captured in these datasets (and see [38] for a discussion of the issues in classifying true opioid use from administrative data.) The cutoff of 90 days of opioid prescription during a calendar year was selected to identify individuals with long-term use, but could instead mark several noncontiguous episodes of acute pain, and 90 days is still only one quarter of the whole year. The pharmacy data were not analyzed to determine daily vs episodic use. Temporally overlapping prescriptions were counted as part of a single episode, and may have omitted some long-term users. The current algorithm for determining long-term use was thus assumed to be conservative, but there were no obvious secular trends that would have changed the categorization over the course of the study. The chronic pain sample was identified through diagnostic codes, and may not be inclusive or representative of all patients who used opioids for chronic pain. All of these limitations would apply equally to all the years of the study, and are thus assumed not to modify significantly the estimates of trends in prescriptions over time.

Conclusion

In two large health systems, the prevalence of long-term opioid prescription increased consistently from 2000 to 2005 in all age and sex groups, but has not reached an epidemic level. The vast majority of patients with chronic pain do not receive long-term opioids. The prevalence of long-term opioid prescription varied considerably between a large integrated health plan and a state Medicaid plan, between men and women, and between younger and older patients, suggesting that these groups used opioid medications differently, but all groups showed similar increases over time. In HealthCore, a large integrated health plan, women aged 65 and over had the highest prevalence of long-term opioid prescription, and in Arkansas Medicaid, middle-aged women and men had the highest prevalence. No single cause accounts for the observed increase in long-term opioid prescription, but the most likely explanations are the increased emphasis on pain control and analgesia during the last decade, and increased patient demand for opioids. Further work is merited to determine the broader medical and social consequences of increased use of opioids in different age and sex groups. When prescribing opioid medications, providers should be aware of the specific needs of their patients as well as the potential risks of their use.