Thursday, March 3, 2011

Biased UDT Research Distorts Opioid Misuse, Again

UDT As an aid for monitoring patient responsiveness to opioid analgesic therapy, urine drug testing (UDT) can be useful in providing better pain care. However, the field is becoming fraught with questionable research studies sounding alarms about opioid misuse detected by UDT. Such studies, often supported by UDT laboratories, seem more like biased infomercials than valid clinical investigations, and a recently published report provides yet another example of this.

The investigation in question assessed the health economic burden imposed by long-term opioid analgesic users and whether nonadherence with prescribed opioid regimens, as determined by UDT, increases healthcare costs in these patients [Leider et al. 2011]. The article appeared in the January 2011 edition of the American Journal of Managed Care and was sponsored by Ameritox, a drug-testing services laboratory.

Data for this retrospective study were obtained from a large managed-care claims database in the United States covering July 1, 2005 through September 20, 2008. Twelve month healthcare utilization and costs were compared for matched groups of long-term opioid users with pain versus non-opioid users without chronic pain (n = 49,425 in each group). Long-term, or chronic, prescription-opioid use was defined as at least a 120-day supply taken during a 6 month period. A subset of long-term opioid users who had undergone monitoring via urine drug testing (UDT) was identified (n = 2,100), and comparisons were made between patients in this cohort who were identified as either being adherent versus “likely nonadherent” with their opioid regimens based on the UDT assays.

The authors found that, during the study period, long-term opioid users had significantly greater 1-year healthcare utilization and costs than matched nonusers (means $23,049 vs $4,975; p <0.001). Among the drug-tested cohort of opioid users, 79% overall were identified as being likely nonadherent with their regimen due to the following: (a) higher than expected concentrations of prescribed opioid = 47% of subjects, (b) lower than expected amounts of prescribed opioid = 16%, (c) no prescribed opioid = 43%, (d) unprescribed controlled substances = 41%, (e) illegal drugs = 12%. [Patients could be nonadherent in multiple categories so numbers add up to more than 100%.]

Adherent patients (n = 442) had significantly lower total healthcare costs than nonadherent patients (n = 1,658; means $23,160 vs $26,433; p = .036). Multivariate analyses showed that the presence of higher than expected levels of the prescribed opioid was most prominently associated with elevated healthcare costs; curiously, however, the presence of illegal drugs was significantly associated with lower costs.

The authors conclude that long-term opioid users represent a substantial cost burden relative to similar patients without evidence of chronic pain and not taking opioids. Nonadherence to the long-term opioid regimen incurs greater healthcare costs than among adherent opioid users, and opioid overuse fosters even more elevated costs. They propose that UDT can identify patients who are likely to be nonadherent, and improving adherence could help to reduce healthcare costs. In particular, they stress, “patients with urine drug levels that were higher than expected using a proprietary algorithm [ie, Ameritox Rx Guardian(sm) system specified in the article] were predicted to have significantly higher costs than patients whose test results were within an expected range.”

COMMENTS & CAVEATS: We have previously discussed biased published research along these lines authored by employees of Millennium Laboratories [UPDATE here], which found unexpectedly high therapeutic noncompliance rates among 20,500 patients monitored by UDT. And, we critiqued an earlier troublesome UDT study sponsored and reported by Ameritox in 2009 [UPDATE here], which happened to find opioid-noncompliance rates among 700,000 patients tested quite similar to those in the present study.

This new study by Leider and colleagues [2011], continues to promote the allegation that patients with chronic pain taking long-term opioids will most likely NOT follow the prescribed regimen, so UDT is essential for detecting and stemming such abuses. Plus, it adds another layer of controversy by linking opioid therapy and nonadherence with such therapy to significantly elevated burdens of healthcare utilization and costs. However, there are many troubling aspects of this study worth closer examination in some detail…
  • For starters, of the 5 listed authors, the first 2 are employees of Ameritox and the other 3 work for the firm contracted by Ameritox to conduct the study. Although these conflicts of interest are duly acknowledged at the end of the paper, this is an archetypal example of potential self-serving bias that cannot be ignored.

  • Perhaps most astounding is that nearly 8 out of 10 patients undergoing UDT were detected as being nonadherent with their prescribed opioid regimens. The similarity in nonadherence rates with the earlier study in 2009 are curious, and the 79% nonadherence rate in this present study might be most representative of a population comprised exclusively of patients at high risk for such deviance. One might suspect that either (a) the UDT assays were oversensitive or inaccurate, (b) the definitions of nonadherence were too inclusive or inappropriate, (c) the drug-tested population under study was unique in certain aspects that were not reported. Here are additional considerations that make the nonadherence rates suspect:

    • Assessments of nonadherence due to higher or lower than expected opioid concentrations relied upon a proprietary quantitative-testing algorithm exclusive to Ameritox, called Rx Guardian. There have been few published studies describing its use and, because it is proprietary (ie, “secret”), independent and unbiased research to verify its accuracy, sensitivity, and specificity has not been feasible. In the past, the use of quantitative assays based on urinalysis have been controversial due to variations in individual patient pharmacokinetics, pharmacodynamics, and pharmacogenetics. Therefore, the evidence in this current study must either be blindly accepted on faith or set aside until such time as the validity and reliability of the testing methods are more objectively determined.

    • The reported incidences of excess opioid (47%), deficient opioid (16%), or no opioid (43%) — which seem mutually exclusive and should not add up to more than 100% as they do — can sometimes be explained by individual patient metabolism or interactions with comedications, which has nothing to do with nonadherence.

    • Even in cases where patients actually may have been under- or over-using their prescribed opioids, this research makes no attempt to account for motivations. Were dosages misprescribed in the first place, leading patients to increase or decrease the dose on their own (or, perhaps, instructed to do so by the prescriber without a note of such being made in the record)? Were patients prescribed supplemental analgesics by another healthcare provider? Such data are not captured in the analyses.

    • Regarding the presence of illicit drugs, the authors do not specify which drugs were involved in this category of nonadherence or their percentages. Past research has shown that marijuana is most frequently implicated as an illicit drug in such cases, and no important distinctions are made between medicinal use (legal in certain states) to enhance pain relief versus purely recreational use.

    • In cases of non-prescribed drugs being present, it is unknown if these might have been prescribed by healthcare providers unknown to the practitioner ordering the UDT.

    • The data do not indicate what percentage of UDTs were ordered due to suspected risky or aberrant behaviors; so, it cannot be assumed that nonadherence was found in a typical patient cohort with chronic pain.

    • Finally, an earlier-reported examination of 281 patients with chronic pain [UPDATE here] found that nearly half of the subjects (48%) exhibited nonadherence to their opioid regimens, with only 14% admitting overuse and 34% reporting underuse of their medication, and such activity was only an occasional occurrence in many patients. Although the data were not confirmed via UDT, the striking contrasts with nonadherence rates reported by Leider and colleagues are noteworthy.

  • As for healthcare utilization and cost differences between groups, long-term opioid users were obviously sicker than nonusers; the authors note that 18 of 20 common comorbidities often associated with chronic pain occurred more frequently in opioid users. Consequently, it is not surprising that the overall healthcare utilization and costs were higher for opioid users. Along with that, however, while differences between the groups compared are statistically significant, due to high variability within groups they are small or modest in terms of effect size. More specifically…

    • Most of the difference in healthcare utilization between groups was associated with ambulatory care visits, which has a very large effect size (Cohen’s d = 0.95 [which is not calculated in the study report; see *note below for an explanation]). Whereas, effect sizes for emergency care and hospital admissions were 0.26 and 0.34, respectively; which signify small and probably unimportant differences between groups. It might be expected that patients with chronic pain would visit their healthcare providers more frequently, whether or not they are being treated with opioids.

    • For healthcare costs, effect size differences between opioid users and nonusers are only modest overall (d = 0.57) though statistically significant (p<0.001), and differences between adherent vs nonadherent opioid users are not meaningful (d = 0.11), although this also was reported in the study as significant (p=0.036).

    • The generally modest or small effect sizes in costs — even though there are differences between groups that are statistically significant — came about because (a) there were sufficient numbers of subjects being studied to detect small differences as being statistically significant (ie, there was adequate statistical “power”), and (b) there appeared to be broad ranges (ie, high variability) in data within each group for most measurements. Considering that the opioid-treated patients had much greater morbidity, as would be expected, it is surprising that the differences in costs are not more robust. And, it appears that adherence with the opioid regimen really does not make any important difference in costs.

    • The suggestion that higher than expected opioid levels incurred greater costs needs further examination. If true, this could represent a valid use of UDT to help detect if a patient is not metabolizing the prescribed opioid as expected, hence they are seeking more healthcare services for unrelieved pain or troublesome side effects. Or, patients might have been taking more opioid due to inadequately prescribed dosing and they are seeking added care to resolve the issue. Still, the finding that these problems may have occurred in nearly half of the patients (47%) seems remarkable.

    • It is interesting that illicit-drug users incurred significantly lower healthcare costs. The authors speculate this may be because such persons neglect their health and, therefore, are less likely to seek healthcare services. Or, possibly, their pain is actually fictitious to begin with. However, an alternate explanation is that their use of illicit drugs with analgesic potential, particularly marijuana, decreases their need for healthcare services relating to unresolved pain. Accurate answers cannot be determined due to a lack of data in this report.

  • Finally, as we have previously noted [here], data-mining approaches such as used in this retrospective study — ie, taking information selectively from a large database that was not originally designed for the purposes of the research at hand — are prone to many types of error. This is largely because critical information often is missing, such as possible factors that might have contributed to detected opioid nonadherence (as noted in the points above), and this tends to invalidate the clinical significance of such studies.
For the above reasons, we must question the validity of this study’s methodology and the usefulness of its conclusions. The extended discussion above was provided because there can be serious concerns if the implications of this sort of research are not put into proper perspective. Leider et al. state, “The likely nonadherence classification is not synonymous with substance abuse,” but they also add, “… certain types of nonadherence could suggest abuse or misuse of controlled or illicit drugs.” Despite these mildly qualifying remarks, there is a presumption throughout the study that “likely nonadherence” is due to intentional misbehaviors of some sort by most patients prescribed long-term opioids.

A danger is that, based on publication of studies like this, as well as the earlier studies noted above, future writers, researchers, or government agencies will cite the studies as prior evidence to support claims of gross nonadherence to long-term opioid regimens by patients with chronic pain. And, this present study might be used to convey the added onus of greater burdens placed on healthcare resources by such patients as an economic justification for instituting universal and frequent UDT. Such proposals are advantageous for UDT laboratories but would be erroneous based on our assessments that look beyond the study abstracts; however, other reviewers, with their own agendas, may not look at the reports in such depth or be interested in more critical assessments.

We would be interested to know what readers think, based on their own observations or assessments of these studies on UDT.

> Leider HL, Dhaliwal J, Davis EJ, Kulakodlu M, Buikema AR. Healthcare Costs and Nonadherence Among Chronic Opioid Users. Am J Manag Care. 2011;17(1):32-40 [
article and appendices here].
*For those curious about Effect Size, this is a measure of the strength of an association between two variables, such as the differences between means (in the above study). Effect size conveys a qualitative estimate of the importance of the association, and it often helps interpret the practical meaning of other statistics such as p-values. “Cohen's d,” a commonly applied measure of effect size, indicates the amount of difference between two groups in standard units and suggests how big or small a significant difference really is in more clinically meaningful terms. As numbers of subjects in study groups (the ‘n’ for each group) are increased, more statistical power is achieved for being able to detect small effect sizes. An effect size (d) between two means within a range encompassing 0.20 is considered small (possibly clinically non-significant), 0.50 is medium, and 0.80 is large. Differences between means with small-to-medium effect sizes can be statistically significant (eg, p≤0.05) yet be clinically unimportant. To validly detect a small effect size between two group means at a p=0.05 significance level requires about 400 subjects in each group; larger effect sizes can be validly detected with fewer enrolled subjects. An easy-to-use effect size calculator (from the University of Colorado) when the mean values and their standard deviations are known can be found [
here]. For more information on statistical power and effect sizes, see this classic article: Cohen J. A Power Primer. Psych Bull. 1992;112(1):155-159 [article PDF here].